Method for cooperative transmission of multiple video streams based on dynamic priority

Abstract

The present application relates to the technical field of resource allocation, in particular to a multi-path video stream cooperative transmission method based on dynamic priority, which constructs a priority allocation model for mode recognition to generate real-time task priority by collecting task context characteristic parameters, system resource state, user behavior response and computing node load data in real time. A task scheduling strategy generation model is established, a priority weighted performance evaluation function, a memory-computing unit occupancy rate prediction matrix and a gradient optimization target under adjusted resource constraints are used for resource allocation to form an initial scheduling strategy. Through multi-dimensional parameter fusion analysis of data by an adaptive optimization decision model, a resource reallocation correction vector is generated, an online iterative optimization mechanism is used to dynamically adjust the initial scheduling strategy, and a final real-time optimized task scheduling strategy is output. The closed-loop optimization of dynamic evaluation of task priority and resource allocation is realized, and efficient cooperative execution of multi-path video processing tasks is ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of resource allocation, and particularly to a multi-path video stream cooperative transmission method based on dynamic priority. BACKGROUND

[0002] With the rapid development of general computing and digital information processing technology, complex application systems that need to process multiple real-time and interactive data streams concurrently are increasingly popular, such as remote collaboration platforms that require low-latency responses, real-time monitoring and analysis systems, and multi-user interactive simulation environments. These systems are usually run on general-purpose digital data processing systems, and their performance is highly dependent on effective management of internal computing resources.

[0003] Currently, in order to achieve efficient and orderly processing of these heterogeneous data streams, existing digital data processing systems usually employ a series of computing resource scheduling strategies (such as fixed priority-based or round-robin scheduling).

[0004] However, in the face of dynamic and highly task load-varying application scenarios, existing technologies still have significant bottlenecks. First, in the case of dynamic changes or high competition for system computing resources, task scheduling and resource allocation algorithms are often application-independent, making it difficult to distinguish and prioritize different processing tasks (and their associated data streams) according to their real-time importance under specific application logic. Second, existing system-level priority management mechanisms are usually decoupled from high-level application state information, lacking mechanisms for real-time sensing and integrating this information. This results in the system scheduler being unable to utilize this valuable contextual information to dynamically and finely adjust the relative priorities of concurrently executed tasks. Third, existing methods need to be improved in terms of the efficiency of concurrent task cooperative execution. When a processing system needs to manage and execute a large number of data processing jobs simultaneously, the inability to intelligently allocate resources and control execution flow based on dynamic priority differences between tasks can lead to delays in critical tasks, affecting the overall processing throughput and response speed of the system.

[0005] Therefore, how to design a dynamic task priority evaluation model that deeply integrates application layer context information, and implement an intelligent and cooperative computing resource management and task scheduling control method based on this model at the operating system or system software level, to ensure the timely completion of critical processing tasks, optimize system resource utilization, and improve the overall performance (such as throughput and responsiveness) of the system, has become a technical problem that needs to be solved in the field of digital data processing system design and operating system optimization. SUMMARY

[0006] The application aims to provide a multi-path video stream cooperative transmission method based on dynamic priority, which constructs a priority allocation model to generate real-time task priority by collecting task context characteristic parameters, system resource state, user behavior response and computing node load data in real time, establishes a task scheduling strategy generation model, adopts a priority weighted performance evaluation function, a memory-computing unit occupancy rate prediction matrix and a resource constraint gradient optimization target to perform resource allocation, and forms an initial scheduling strategy.

[0007] To achieve the above object, the application provides the following technical scheme.

[0008] The multi-path video stream cooperative transmission method based on dynamic priority comprises the following steps.

[0009] Real-time collection of task context characteristic parameters, system resource state parameters, user behavior response data and computing node load parameters associated with multi-path video data processing tasks, wherein the task context characteristic parameters comprise user operation focus coordinates, application event trigger sequences and a predefined task rule set;

[0010] Construction of a priority allocation model for pattern recognition of the task context characteristic parameters to generate real-time priority of each video data processing task;

[0011] Establishment of a task scheduling strategy generation model, resource allocation for the multi-path video data processing tasks based on the real-time priority to generate an initial scheduling strategy; the construction method of the task scheduling strategy generation model comprises generation of a priority weighted performance evaluation function, a predicted memory-computing unit occupancy rate matrix and a gradient optimization target under adjusted resource constraints;

[0012] Construction of an adaptive optimization decision model for multi-dimensional parameter fusion analysis of the system resource state parameters, user behavior response data and computing node load parameters to generate a resource reallocation correction vector; online iterative optimization of the initial scheduling strategy according to the resource reallocation correction vector to generate a final real-time optimized task scheduling strategy.

[0013] Preferably, the construction of the priority allocation model comprises a feature extraction fusion layer, a pattern recognition layer and a priority mapping layer.

[0014] The feature extraction and fusion layer pre-processes the task context feature parameters, extracts spatial-temporal features from the user operation focus coordinates, extracts time sequence pattern features from the application event trigger sequence, and extracts semantic features from the predefined task rule set, and generates comprehensive task context state features by fusing the spatial-temporal features, the time sequence pattern features, and the semantic features; the pattern recognition layer classifies and clusters the fused comprehensive task context state features to recognize task patterns; and the priority mapping layer maps the task patterns to priority scores according to a mapping function to generate real-time priorities of the video data processing tasks.

[0015] Preferably, the priority mapping layer generates the real-time priorities of the video data processing tasks by the following process:

[0016] obtaining the task patterns recognized and output by the pattern recognition layer; calling a mapping function, which is a corresponding relationship between the task patterns and the priority scores; the mapping function is obtained by using a machine learning model to fit the relationship between the task patterns and the priority scores; inputting the obtained task patterns into the mapping function for calculation, and assigning a corresponding priority score to each task pattern to generate the real-time priorities of the video data processing tasks.

[0017] Preferably, the task scheduling strategy generation model includes a resource demand evaluation layer, a resource demand prediction layer, and an optimization target adjustment layer.

[0018] The resource demand evaluation layer obtains the weighted performance expected value of each video data processing task according to the real-time priorities and the task patterns, and in combination with the priority weighted performance evaluation function; the resource demand prediction layer predicts the predicted resource demand amount required by each video data processing task in a future period based on historical task context feature parameters and using the predicted memory-computing unit occupancy rate matrix by using a scheduling algorithm; the predicted resource demand amount includes memory and computing unit resource amounts; and the optimization target adjustment layer takes maximizing the overall weighted performance expected value as a target while satisfying the adjustment resource constraint condition, and based on the predicted resource demand amount, uses gradient optimization to solve the processing resource allocation scheme of each task to generate the initial scheduling strategy, which includes the computing unit proportion, memory size, and processing time slice allocated to each task.

[0019] Preferably, the generation process of the priority-weighted performance evaluation function comprises: a basic performance measurement standard for quantifying the processing efficiency of a single-path video data processing task under the same resource and the contribution degree to user experience; an association model of real-time priority and basic performance quantity is generated by integrating the basic performance measurement standard and real-time priority; the association model is solidified as a calculation rule, an expected value of weighted performance is output, and parameters in the function are set according to an optimization target and experimental data;

[0020] The process of generating the predicted memory-computing unit occupancy rate matrix comprises: obtaining historical data of each path video data processing task in a running process, including the actual allocated memory and computing unit resource quantity in a specific time period and the actual generated memory occupancy peak value, average value and computing unit usage rate under corresponding resource allocation; based on the historical data, a statistical analysis is applied to identify and establish a quantitative prediction relationship model between a task context feature parameter combination and future memory and computing unit resource occupancy rate; the result of the prediction relationship model is analyzed to form the predicted memory-computing unit occupancy rate matrix.

[0021] Preferably, the adaptive optimization decision model comprises a state evaluation layer and an adjustment strategy generation layer.

[0022] The state evaluation layer receives the system resource state parameters, user behavior response data and computing node load parameters in real time; a multi-dimensional parameter fusion technology is adopted to comprehensively analyze the input real-time data, generate a multi-path video stream cooperative transmission vector and identify potential risk points; the adjustment strategy generation layer generates the resource reallocation correction vector according to the potential risk points, and the resource reallocation correction vector comprises parameters for explicitly indicating the need to increase and / or decrease the amount of computing resources, memory quota and network bandwidth priority allocated to specific high-priority and / or low-priority video data processing tasks and adjusting global parameters.

[0023] Preferably, the process of generating the resource reallocation correction vector specifically comprises: applying the resource reallocation correction vector to the initial scheduling strategy; adjusting the corresponding resource allocation parameters in the initial scheduling strategy according to the adjustment direction and amplitude indicated by each component in the resource reallocation correction vector; in the adjustment process, it is checked in real time whether the adjusted resource allocation exceeds the total quantity constraint of existing resources; if the constraint is exceeded, rebalancing is performed according to a preset rule; the online iterative optimization is repeated to execute all steps in each scheduling period or a predetermined time interval, and the scheduling strategy of the last period is adjusted based on the latest resource reallocation correction vector to output the optimal real-time task scheduling strategy in the current period.

[0024] Compared with the prior art, the application has the following beneficial effects:

[0025] 1、The application is based on a dynamic priority allocation and cooperative scheduling method, which realizes multi-dimensional information fusion and accurate resource allocation by collecting task context features, resource states and user behavior data in real time. The method constructs a feature extraction and fusion layer, a pattern recognition layer and a priority mapping layer, and automatically fits the relationship between task patterns and priority scores using machine learning, effectively overcoming the defects of rigid resource configuration and long response delay in traditional static scheduling schemes. The video data processing efficiency is greatly improved, the scheduling complexity is reduced, accurate grading and resource optimization are realized, the transmission process is ensured to be efficient and reliable, and the user experience and overall efficiency are optimized.

[0026] 2、The application introduces an adaptive optimization decision model to realize online resource reallocation and scheduling strategy iteration optimization. The model analyzes resource states, user behavior feedback and computing node load, identifies potential risks using multi-dimensional parameter fusion technology, and generates a resource reallocation correction vector to realize dynamic resource adjustment of each video task. According to the real-time collected data, the scheduling strategy is automatically adjusted, effectively alleviating the bottleneck problem caused by resource shortage, ensuring the stability and anti-interference ability of the video transmission process, and significantly improving the overall network cooperative performance and operation reliability.

[0027] 3、The application comprehensively uses statistical analysis, multi-dimensional parameter fusion and gradient optimization algorithm to construct a prediction memory and computing unit occupancy rate matrix, realizes task resource demand quantity quantitative prediction. By real-time monitoring of historical operation data, based on the optimization target of priority weighted performance evaluation function, the initial scheduling strategy is automatically generated and iteratively corrected online, ensuring that each task obtains the optimal allocation under the specified resource constraints. Not only improves the resource utilization efficiency, but also can be continuously self-adapted according to the dynamic changes, ensures the network load balancing and the improvement of overall performance. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 The application provides a multi-channel video stream cooperative transmission method flowchart based on dynamic priority;

[0029] Figure 2 The application provides a priority allocation model structure schematic diagram;

[0030] Figure 3 The application provides a task scheduling strategy generation model structure schematic diagram;

[0031] Figure 4 The application provides a dynamic priority cooperative transmission flowchart. DETAILED DESCRIPTION

[0032] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of the present application.

[0033] With the rapid development of network communication technology and multimedia processing technology, real-time video interactive systems based on networks have been widely applied in many fields such as remote education, video conference, remote medical treatment, collaborative design, etc. In particular, in the scenes of remote interactive teaching and online classroom, in order to simulate a real teaching environment and enhance interactivity, video streams from multiple sources often need to be simultaneously collected, transmitted, processed and presented.

[0034] However, the existing multi-stream video stream transmission method still has significant technical bottlenecks when dealing with complex interactive scenes. First, in the case of limited or fluctuating network bandwidth resources, traditional bandwidth allocation and quality adjustment strategies (such as average allocation, priority based on simple rules) often fail to effectively distinguish the actual importance of different video streams in a specific application context. Second, the priority management mechanism of the existing technology is usually static or too simplified, lacking deep perception of the specific business logic and real-time interactive state of the upper-layer application. Third, the existing method needs to be improved in terms of multi-stream collaborative transmission efficiency. When a large number of video streams need to be transmitted and received simultaneously, the system tends to adjust each stream independently or based on the overall network condition, failing to fully utilize the priority differences between streams for collaborative optimization. Therefore, how to design a method that can perceive application scenarios and user interaction intentions, dynamically determine the transmission priority of multi-stream video streams, and based on this priority, intelligently and collaboratively allocate resources and control transmission, to improve the transmission quality of key video streams and overall user experience under limited network resources, has become a technical problem to be solved in the field of multi-stream video stream transmission technology.

[0035] Embodiment one:

[0036] Please refer to Figure 1 The present application provides a multi-stream video stream collaborative transmission method based on dynamic priority, and the technical solution is as follows:

[0037] Real-time collection of task context feature parameters, system resource state parameters, user behavior response data and computing node load parameters associated with multi-stream video data processing tasks, the task context feature parameters including user operation focus coordinates, application event trigger sequence and predefined task rule set;

[0038] The priority allocation model is constructed to perform mode recognition on the task context feature parameters, and real-time priorities of the video data processing tasks are generated.

[0039] Further, the priority allocation model comprises a feature extraction fusion layer, a mode recognition layer, and a priority mapping layer, as shown in Figure 2 ;

[0040] The feature extraction fusion layer pre-processes the task context feature parameters, extracts space-time features of the user operation focus coordinates, extracts time sequence mode features of the application event trigger sequence, and extracts semantic features of the predefined task rule set. The space-time features, the time sequence mode features, and the semantic features are fused to generate comprehensive task context state features. The mode recognition layer classifies and clusters the fused comprehensive task context state features to recognize task modes. The priority mapping layer maps the task modes to priority scores according to a mapping function to generate real-time priorities of the video data processing tasks.

[0041] In this embodiment, the task context feature parameters, the system resource state parameters, the user behavior response data, and the computing node load parameters associated with the multi-channel video data processing tasks are collected in real time, and the priority allocation model is constructed to perform mode recognition on the task context feature parameters, and real-time priorities of the video data processing tasks are generated. This method can comprehensively perceive the system running environment and the user demand, and provide rich data basis for subsequent priority allocation and resource scheduling. This multi-source data collection mechanism can dynamically adapt to changing application scenarios and user behaviors, and improves the intelligent level and response ability.

[0042] Further, the process of the priority mapping layer generating real-time priorities of the video data processing tasks comprises:

[0043] The task modes recognized and output by the mode recognition layer are obtained. A mapping function is called, which is the corresponding relationship between the task modes and the priority scores. The mapping function is obtained by using a machine learning model to fit the relationship between the task modes and the priority scores. The obtained task modes are input into the mapping function for calculation, a corresponding priority score is allocated to each task mode, and real-time priorities of the video data processing tasks are generated.

[0044] In this embodiment, the priority mapping layer automatically fits the mapping relationship between the task mode and the priority score by using a machine learning model, realizing the dynamic generation of the task priority. The limitations of traditional static priority allocation are overcome, and the priority can be quickly adjusted according to the changes of the real-time task mode, ensuring that the key tasks are given priority in complex and variable application scenarios. Combined with the real-time collected multi-dimensional data and the priority allocation model, this method can accurately identify the importance and urgency of the task, providing a scientific basis for subsequent resource allocation and scheduling, and significantly improving the response speed and processing efficiency of key tasks.

[0045] A task scheduling strategy generation model is established, and the multi-path video data processing tasks are processed based on the real-time priority to allocate processing resources and generate an initial scheduling strategy. The construction method of the task scheduling strategy generation model includes generating a priority weighted performance evaluation function, a predicted memory-computing unit occupancy rate matrix, and a gradient optimization target under an adjusted resource constraint condition.

[0046] Further, the task scheduling strategy generation model includes a resource demand evaluation layer, a resource demand prediction layer, and an optimization target adjustment layer, as described in Figure 3 ;

[0047] The resource demand evaluation layer obtains the weighted performance expected value of each video data processing task according to the real-time priority and task mode, combined with the priority weighted performance evaluation function. The resource demand prediction layer predicts the predicted resource demand amount required by each video data processing task in a future period based on historical task context feature parameters and the predicted memory-computing unit occupancy rate matrix by using a scheduling algorithm. The predicted resource demand amount includes memory and computing unit resource amount. The optimization target adjustment layer aims to maximize the overall weighted performance expected value while meeting the adjusted resource constraint condition, and based on the predicted resource demand amount, uses gradient optimization to solve the processing resource allocation scheme of each task to generate the initial scheduling strategy, which includes the computing unit proportion, memory size, and processing time slice allocated to each task.

[0048] In this embodiment, the task scheduling strategy generation model generates an initial scheduling strategy based on real-time priority and task mode through the collaborative work of the resource demand evaluation layer, the resource demand prediction layer, and the optimization target adjustment layer. The priority and resource demand of the task are comprehensively considered, and the gradient optimization algorithm is used to allocate resources under the resource constraint condition, realizing the refinement and scientization of resource allocation. On the basis of real-time data collection and dynamic generation of priority, the priority is further converted into a specific resource allocation scheme through the task scheduling strategy generation model, ensuring that high-priority tasks are given more resource support, thereby improving the collaborative execution efficiency of multi-path video processing tasks and the overall performance of the system.

[0049] Further, the generation process of the priority-weighted performance evaluation function includes: a basic performance measurement standard for quantifying the processing efficiency and contribution degree to user experience of a single-path video data processing task under the same resource; an association model of real-time priority and basic performance quantity is generated based on the basic performance measurement standard and real-time priority; the association model is solidified as a calculation rule, an expected value of weighted performance is output, and parameters in the function are set according to optimization targets and experimental data;

[0050] The process of generating the predicted memory-computing unit occupancy rate matrix includes: obtaining historical data of each path of video data processing tasks in the running process, including the actual allocated memory and computing unit resource quantity in a specific time period and the actual generated memory occupancy peak value, average value and computing unit usage rate under corresponding resource allocation; based on the historical data, statistical analysis is applied to identify and establish a quantitative prediction relationship model between the task context feature parameter combination and future memory and computing unit resource occupancy rate; the result of the prediction relationship model is analyzed to form the predicted memory-computing unit occupancy rate matrix.

[0051] In the embodiment, by constructing the priority-weighted performance evaluation function and the predicted memory-computing unit occupancy rate matrix, the quantitative evaluation of task processing effect and resource demand is realized. The priority-weighted performance evaluation function combines the basic performance measurement and real-time priority to ensure that the resource allocation is highly matched with the task importance; the prediction matrix accurately predicts the future resource occupancy rate through historical data analysis, providing a scientific basis for resource allocation. On the basis of priority allocation and task scheduling strategy generation, the performance evaluation and resource demand prediction mechanism are further introduced to realize the dynamic optimization and forward-looking adjustment of resource allocation, significantly improving the resource utilization efficiency and stability of task execution.

[0052] The adaptive optimization decision model constructs a multi-dimensional parameter fusion analysis on the system resource state parameters, user behavior response data and computing node load parameters to generate a resource reallocation correction vector; and performs online iterative optimization on the initial scheduling strategy according to the resource reallocation correction vector to generate a final real-time optimized task scheduling strategy.

[0053] Further, the adaptive optimization decision model includes a state evaluation layer and an adjustment strategy generation layer.

[0054] The state evaluation layer receives the system resource state parameters, user behavior response data and computing node load parameters in real time; adopts a multi-dimensional parameter fusion technology to comprehensively analyze the input real-time data, generates a multi-path video stream cooperative transmission vector, and identifies potential risk points; the adjustment strategy generation layer generates the resource reallocation correction vector according to the potential risk points, and the resource reallocation correction vector includes parameters such as the amount of computing resources, memory quota, network bandwidth priority and global adjustment priority that need to be increased and / or decreased for specific high-priority and / or low-priority video data processing tasks.

[0055] In this embodiment, the adaptive optimization decision model can realize real-time perception of system running state and user behavior changes and generate a resource reallocation correction vector through multi-dimensional parameter fusion analysis of the state evaluation layer and the adjustment strategy generation layer. The vector guides dynamic adjustment of resources for high-priority and low-priority tasks, ensuring that critical tasks are prioritized for support when resources are scarce. On the basis of priority allocation, task scheduling and resource demand prediction, an adaptive optimization mechanism is further introduced to realize real-time adjustment and optimization of resource allocation, significantly improving adaptability and stability in complex network environments.

[0056] Further, the process of generating the resource reallocation correction vector specifically includes: applying the resource reallocation correction vector to the initial scheduling strategy; adjusting the corresponding resource allocation parameters in the initial scheduling strategy according to the adjustment direction and amplitude indicated by each component in the resource reallocation correction vector; in the adjustment process, real-time check is performed on whether the adjusted resource allocation exceeds the total amount constraint of existing resources; if the constraint is exceeded, rebalancing is performed according to a preset rule to ensure the feasibility of resource allocation; the online iterative optimization repeatedly executes all steps in each scheduling period or a predetermined time interval, and adjusts the scheduling strategy of the previous period based on the latest resource reallocation correction vector, realizes continuous optimization of the scheduling strategy with dynamic changes in system state and user behavior, and outputs the optimal real-time task scheduling strategy in the current period.

[0057] In this embodiment, through the online iterative optimization mechanism, the system state and user behavior can be continuously monitored, and the scheduling strategy can be dynamically adjusted according to the latest resource reallocation correction vector. This ensures that resource allocation can quickly adapt to real-time changes, and realizes continuous optimization of the scheduling strategy. On the basis of real-time data collection, dynamic priority generation, task scheduling, performance evaluation and adaptive optimization, closed-loop control is realized through online iterative optimization, ensuring the stability and efficiency of multi-path video stream cooperative transmission, and significantly improving overall performance and user experience.

[0058] The application realizes forward-looking resource allocation through a prediction matrix and gradient optimization, significantly improves resource utilization efficiency and key task guarantee capability compared with the average allocation mode of round-robin scheduling, and specific reference is made to Table 1.

[0059] Table 1 Resource allocation efficiency comparison table

[0060] Comparison index Round-robin scheduling algorithm Invention scheme Lifting amplitude / Explanation Resource allocation response time Fixed cycle round-robin, slow response Online iterative optimization, real-time adjustment of resource allocation strategy Response time is shortened by 70% Resource waste rate High (average allocation leads to low efficiency) Fine allocation based on performance evaluation and gradient optimization Waste rate is reduced by 55% High-priority task guarantee Unable to distinguish priority Weighted performance evaluation function ensures that high-priority tasks get more resources Key task resource satisfaction rate increased by 85%

[0061] Embodiment one provides a dynamic priority-based multi-path video stream cooperative transmission method, which realizes the deep integration of task priority evaluation and resource allocation by real-time acquisition of multi-dimensional data, construction of priority allocation model, generation of task scheduling strategy and adaptive optimization decision. It can dynamically perceive the changes of application scenarios, quickly adjust the resource allocation strategy, and ensure the efficient cooperative execution of multi-path video processing tasks in complex network environment. The application integrates task context perception, dynamic priority evaluation, resource demand prediction and online iterative optimization technologies to build a highly intelligent multi-path video stream cooperative transmission framework. Compared with traditional static scheduling methods, this scheme has significantly improved in resource utilization efficiency, task execution efficiency and system stability. Especially in the scene of limited network bandwidth or severe load fluctuation, this method can preferentially guarantee the quality of key video streams and optimize the overall system performance, providing reliable technical support for multi-user interactive scenarios such as remote collaboration and real-time monitoring, greatly improving user experience and system operation efficiency.

[0062] Embodiment two:

[0063] The dynamic priority-based multi-path video stream cooperative transmission method provided by the application, the specific flow chart is referred to Figure 4 , and the specific scheme is as follows:

[0064] Real-time acquisition of task context feature parameters, system resource state parameters, user behavior response data and computing node load parameters associated with multi-path video data processing tasks, the task context feature parameters include user operation focus coordinates, application event trigger sequence and pre-defined task rule set;

[0065] The user operation focus coordinates record the operation focus position and timestamp t of the user in real time through the input device (such as mouse, touch screen), form a time series data sequence, and reflect the user's current focus point;

[0066] The application event trigger sequence records the events in the application, such as button click, video switching, etc., forms an event sequence, and each event contains event type (such as "play", "pause") and occurrence time;

[0067] The pre-defined task rule set defines priority rules according to application scenarios, for example, "in a video conference, the priority of the speaker video stream is higher than that of the audience video stream", and is stored in the form of rules, such as {rule 1: "speaker priority", rule 2: "high-definition mode priority"}.

[0068] The system resource state parameters include CPU usage, memory occupation, and network bandwidth.

[0069] The CPU usage is obtained by a system API to obtain the overall CPU usage and the usage of each core; the memory occupation records the total memory, used memory, and available memory; and the network bandwidth monitors the upload and download speeds and the bandwidth utilization.

[0070] The user behavior response data includes quality feedback and interaction frequency.

[0071] The quality feedback is the feedback submitted by the user through the interface, such as "video lag" or "audio and video out of sync"; and the interaction frequency is the number of operations of the user within a unit of time, such as 5 mouse clicks and 3 keyboard inputs within 1 minute.

[0072] The computing node load parameters include node CPU load, node memory load, and node network load.

[0073] The priority allocation model is constructed to perform mode recognition on the task context feature parameters to generate real-time priorities of each video data processing task.

[0074] Further, the priority allocation model includes a feature extraction and fusion layer, a mode recognition layer, and a priority mapping layer.

[0075] The feature extraction and fusion layer pre-processes the task context feature parameters, extracts the spatio-temporal features of the user operation focus coordinates, extracts the time sequence mode features of the application event trigger sequence, and extracts the semantic features of the pre-defined task rule set, and generates comprehensive task context state features by fusing the spatio-temporal features, the time sequence mode features, and the semantic features.

[0076] Spatio-temporal features of the user operation focus coordinates:

[0077] Calculation of the focus movement speed: , in pixels / second, where is the position of the user's operation focus at time , and is the position of the user's operation focus at time , and is the horizontal and vertical coordinates under the th feature, with the lower left corner of the screen as the origin, and is the focus movement speed under the th feature.

[0078] Calculate dwell time If the coordinate change is less than a threshold (e.g. 5 pixels) for k consecutive time points , then the dwell time .

[0079] Apply temporal pattern features of event trigger sequence:

[0080] Calculate event frequency using a sliding window (window size w) : .

[0081] Identify patterns: if the event sequence matches a predefined pattern (e.g. "click 3 times in a row"), mark it as a specific pattern.

[0082] Semantic features of predefined task rule set:

[0083] Convert rules to feature vectors, e.g. using one-hot encoding: { "Speaker Priority": [1, 0], "High Definition Mode Priority": [0, 1]}.

[0084] Feature fusion:

[0085] Weighted sum of the above features to form a comprehensive task context state feature vector F:

[0086] ;

[0087] Where, is the weight of the corresponding vector, and R is the rule feature vector.

[0088] The pattern recognition layer classifies and clusters the fused comprehensive task context state feature F to identify task patterns;

[0089] Using the K-means algorithm, F is divided into k clusters, each representing a task mode M (e.g. "high priority interaction mode", "low priority background mode").

[0090] Output the current task mode , Indicates "high priority interaction mode";

[0091] The priority mapping layer maps the task mode to a priority score according to the mapping function, generating real-time priority for each video data processing task.

[0092] Further, the priority mapping layer generates the real-time priority of each video data processing task, including:

[0093] obtaining the task mode identified and output by the mode recognition layer; calling a mapping function, which is a corresponding relationship between the task mode and the priority score; the mapping function uses a machine learning model to fit the relationship between the task mode and the priority score (such as a multilayer perception ); inputting the obtained task mode into the mapping function for calculation, and assigning a corresponding priority score to each task mode , to generate the real-time priority of each video data processing task.

[0094] establishing a task scheduling strategy generation model, allocating processing resources to the multiple video data processing tasks based on the real-time priority, and generating an initial scheduling strategy; the construction method of the task scheduling strategy generation model includes: generating a priority weighted performance evaluation function, a predicted memory-computing unit occupancy rate matrix, and a gradient optimization objective under an adjusted resource constraint condition;

[0095] Further, the task scheduling strategy generation model includes a resource demand evaluation layer, a resource demand prediction layer, and an optimization objective adjustment layer.

[0096] The resource demand evaluation layer obtains the weighted performance expected value of each video data processing task according to the real-time priority and the task mode, in combination with the priority weighted performance evaluation function; the resource demand prediction layer predicts the predicted resource demand amount required by each video data processing task in a future period based on historical task context feature parameters, using the predicted memory-computing unit occupancy rate matrix, by using a scheduling algorithm; the predicted resource demand amount includes memory and computing unit resource amount; the optimization objective adjustment layer takes maximizing the overall weighted performance expected value as the target, while satisfying the adjusted resource constraint condition, and based on the predicted resource demand amount, uses gradient optimization to solve the processing resource allocation scheme of each task, to generate the initial scheduling strategy, which includes the computing unit proportion, memory size, and processing time slice allocated to each task.

[0097] Further, the generation process of the priority weighted performance evaluation function includes a basic performance measurement standard for quantifying the processing efficiency and the contribution degree to user experience of a single video data processing task under the same resource : ; wherein is the throughput of task i, is the delay.

[0098] The basic performance measurement standard and the real-time priority are integrated to generate an association model of the real-time priority to the basic performance amount; the association model is solidified as a calculation rule to output the weighted performance expected value , wherein Real-time priority of task i, and set parameters in the function according to optimization target and experimental data;

[0099] The process of generating the predicted memory-computation unit occupancy matrix comprises: obtaining historical data of each video data processing task in a running process, including the actual allocated memory and computation unit resource amount in a specific time period and the actual generated memory occupancy peak value, average value and computation unit usage rate under corresponding resource allocation; based on the historical data, applying statistical analysis (such as linear regression), identifying and establishing a quantitative prediction relationship model between the task context feature parameter combination and future memory and computation unit resource occupancy rate; analyzing the results of the prediction relationship model to form the predicted memory-computation unit occupancy matrix; wherein the memory occupancy rate and the computation unit occupancy rate can be predicted by linear regression.

[0100] The optimization target adjustment layer aims to maximize the overall weighted performance expectation value while meeting the adjustment resource constraint condition, and based on the predicted resource demand amount, solves the processing resource allocation scheme of each task by using gradient optimization to generate the initial scheduling strategy, and the initial scheduling strategy comprises the computation unit proportion, memory size and processing time slice allocated to each task;

[0101] Maximizing the overall weighted performance expectation value while meeting the resource constraint .

[0102] The resource allocated to task i, is the total resource.

[0103] Solve using gradient descent: , constraint .

[0104] Table 2 Comparison table of system resource utilization

[0105] Resource type Static scheduling utilization rate Invention scheme utilization rate Utilization rate is improved CPU 79% 88% 9% Memory 70% 82% 12% Network bandwidth 50% 68% 18%

[0106] The present application maximizes resource utilization through multi-dimensional parameter fusion and gradient optimization target, especially in the network bandwidth limited scene, and the data specifically refer to Table 2.

[0107] The adaptive optimization decision model is constructed to perform multi-dimensional parameter fusion analysis on the system resource state parameters, user behavior response data and computation node load parameters, and generate a resource reallocation correction vector; the initial scheduling strategy is iteratively optimized online according to the resource reallocation correction vector to generate a final real-time optimization task scheduling strategy.

[0108] Further, the adaptive optimization decision model comprises a state evaluation layer and an adjustment strategy generation layer.

[0109] The state evaluation layer receives the system resource state parameters, user behavior response data and computing node load parameters in real time; adopts a multi-dimensional parameter fusion technology to comprehensively analyze the input real-time data, generates a multi-path video stream cooperative transmission vector, and identifies potential risk points; the adjustment strategy generation layer generates the resource reallocation correction vector according to the potential risk points The resource reallocation correction vector includes parameters that explicitly indicate the amount of computing resources, memory quota, network bandwidth priority and adjustment globality that need to be increased and / or decreased for specific high-priority and / or low-priority video data processing tasks.

[0110] Resource reallocation correction vector The calculation formula is:

[0111] ;

[0112] Wherein, is the current memory occupancy rate, is the current computing unit occupancy rate, and are adjustment coefficients.

[0113] Further, the process of generating the resource reallocation correction vector specifically comprises: applying the resource reallocation correction vector to the initial scheduling strategy; adjusting the corresponding resource allocation parameters in the initial scheduling strategy according to the adjustment direction and amplitude indicated by each component in the resource reallocation correction vector; during the adjustment process, it is checked in real time whether the adjusted resource allocation exceeds the total amount constraint of the existing resources; if it exceeds the constraint, rebalancing is performed according to the preset rule to ensure the feasibility of resource allocation; the online iterative optimization is repeated to execute all steps within each scheduling period or predetermined time interval, and the scheduling strategy of the last period is adjusted based on the latest resource reallocation correction vector, so as to realize the continuous optimization of the scheduling strategy with the dynamic changes of the system state and user behavior, and output the optimal real-time task scheduling strategy in the current period.

[0114] The application significantly improves the user's satisfaction with video smoothness, response speed and multi-task cooperation through dynamic priority and adaptive optimization mechanism.

[0115] Although embodiments of the application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the application, and the scope of the application is defined by the appended claims and their equivalents.

Claims

1. A method for cooperative transmission of multiple video streams based on dynamic priority, characterized in that, The method comprises the following steps: Real-time acquisition of task context characteristic parameters, system resource state parameters, user behavior response data and computing node load parameters associated with multi-channel video data processing tasks, wherein the task context characteristic parameters include user operation focus coordinates, application event trigger sequences and a pre-defined task rule set; Construction of a priority allocation model for pattern recognition of the task context characteristic parameters to generate real-time priorities of each channel of video data processing tasks; The construction of the priority allocation model includes a feature extraction and fusion layer, a pattern recognition layer and a priority mapping layer; The feature extraction and fusion layer pre-processes the task context characteristic parameters, extracts spatio-temporal features from the user operation focus coordinates, extracts time sequence pattern features from the application event trigger sequences and extracts semantic features from the pre-defined task rule set, and generates comprehensive task context state features by fusing the spatio-temporal features, time sequence pattern features and semantic features; the pattern recognition layer classifies and clusters the fused comprehensive task context state features to identify task patterns; and the priority mapping layer maps the task patterns to priority scores according to a mapping function to generate real-time priorities of each channel of video data processing tasks; Establishment of a task scheduling strategy generation model for processing resource allocation of the multi-channel video data processing tasks based on the real-time priorities to generate an initial scheduling strategy; The construction method of the task scheduling strategy generation model includes generation of a priority weighted performance evaluation function, a predicted memory-computing unit occupancy rate matrix and a gradient optimization objective under adjusted resource constraints; The task scheduling strategy generation model includes a resource demand evaluation layer, a resource demand prediction layer and an optimization objective adjustment layer; The resource demand evaluation layer obtains a weighted performance expectation value of each channel of video data processing tasks according to the real-time priorities and task patterns in combination with the priority weighted performance evaluation function; the resource demand prediction layer predicts a predicted resource demand amount required by each channel of video data processing tasks in a future period of time based on historical task context characteristic parameters by using the predicted memory-computing unit occupancy rate matrix by using a scheduling algorithm; the predicted resource demand amount includes memory and computing unit resource amounts; and the optimization objective adjustment layer takes maximizing the overall weighted performance expectation value as an objective while satisfying the adjusted resource constraints, and based on the predicted resource demand amount, solves a processing resource allocation scheme for each task by using gradient optimization to generate the initial scheduling strategy, which includes a computing unit proportion, memory size and processing time slice allocated to each task; Construction of an adaptive optimization decision model for multi-dimensional parameter fusion analysis of the system resource state parameters, user behavior response data and computing node load parameters to generate a resource reallocation correction vector; online iterative optimization of the initial scheduling strategy according to the resource reallocation correction vector to generate a final real-time optimized task scheduling strategy; The adaptive optimization decision model includes a state evaluation layer and an adjustment strategy generation layer. The state evaluation layer receives the system resource state parameters, user behavior response data and computing node load parameters in real time; adopts a multi-dimensional parameter fusion technology to comprehensively analyze the input real-time data, generates a multi-path video stream cooperative transmission vector, and identifies potential risk points; the adjustment strategy generation layer generates the resource reallocation correction vector according to the potential risk points, and the resource reallocation correction vector includes parameters such as the amount of computing resources, memory quota, network bandwidth priority and global adjustment priority that need to be increased and / or decreased for specific high-priority and / or low-priority video data processing tasks.

2. The dynamic priority-based multi-path video stream cooperative transmission method according to claim 1, characterized in that: The priority mapping layer generates real-time priorities of each path of video data processing tasks, including: Obtaining the task mode identified and output by the pattern recognition layer; calling a mapping function, which is a corresponding relationship between the task mode and the priority score; the mapping function is obtained by using a machine learning model to fit the relationship between the task mode and the priority score; inputting the obtained task mode into the mapping function for calculation, assigning a corresponding priority score to each task mode, and generating real-time priorities of each path of video data processing tasks.

3. The dynamic priority-based multi-path video stream cooperative transmission method according to claim 1, characterized in that: The generation process of the priority weighted performance evaluation function includes: a basic performance measurement standard for quantifying the processing efficiency and contribution to user experience of a single path of video data processing tasks under the same resource; generating a real-time priority-basis performance quantity correlation model by integrating the basic performance measurement standard and the real-time priority; solidifying the correlation model as a calculation rule, outputting a weighted performance expected value, and setting parameters in the function according to the optimization target and experimental data; The process of generating the predicted memory-computing unit occupancy rate matrix includes: obtaining historical data of each path of video data processing tasks during the running process, including the actual allocated memory and computing unit resource quantity and the actual generated memory occupancy peak value, average value and computing unit usage rate under corresponding resource allocation; based on the historical data, applying statistical analysis, identifying and establishing a quantitative prediction relationship model between the task context feature parameter combination and the future memory and computing unit resource occupancy rate; analyzing the results of the prediction relationship model to form the predicted memory-computing unit occupancy rate matrix.

4. The dynamic priority-based multi-path video stream cooperative transmission method according to claim 1, characterized in that: The process of generating the resource reallocation correction vector specifically comprises: applying the resource reallocation correction vector to the initial scheduling strategy; adjusting the corresponding resource allocation parameter in the initial scheduling strategy according to the adjustment direction and amplitude indicated in each component of the resource reallocation correction vector; during the adjustment process, checking whether the adjusted resource allocation exceeds the total amount constraint of the existing resources in real time; if the constraint is exceeded, rebalancing is performed according to a preset rule; the online iterative optimization is repeated to execute all steps in each scheduling period or a predetermined time interval, and the scheduling strategy of the previous period is adjusted based on the latest resource reallocation correction vector to output the real-time task scheduling strategy that is optimal in the current period.

Images