Intelligent task unloading system for mobile edge computing in live broadcast platform

Abstract

The invention discloses an intelligent task unloading system for mobile edge computing in a live broadcast platform. The intelligent task unloading system comprises live broadcast platform mobile edge computing equipment, base station computing equipment and a live broadcast platform edge computing server, the live broadcast platform mobile edge computing device comprises a plant community intelligent task unloading module and the like. The base station computing equipment comprises a plant community intelligent task unloading module and the like; the live broadcast platform edge computing server comprises a commodity recommendation computing module, a live broadcast edge computing task computing module and a plant community intelligent task unloading module; the problem of live broadcast edge calculation task unloading in the live broadcast platform is solved by simulating the intelligent behavior of the plant community.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to live broadcast platforms, edge computing and artificial intelligence technologies, and specifically to an intelligent task offloading system for mobile edge computing in live broadcast platforms. Background Art

[0002] In the prior art, the overall architecture of the live broadcast platform generally includes a live broadcast module and a short video module. Custom modules can also be appropriately added according to the needs of the live broadcast platform, including an advertising module and a product recommendation module. The mobile and PC terminals of the live broadcast platform are responsible for collecting video streams and video files, and the live broadcast platform decodes, transcodes, encodes, stores, and forwards video data. Network operators provide the necessary computing resources and storage resources for the network live broadcast platform to complete the processing of live video data.

[0003] With the rapid development of the live broadcast industry, a huge amount of video data has been generated. Traditional cloud servers are far away from the live broadcast source, which can easily lead to high traffic delays. Therefore, low-cost, low-latency edge computing has also begun to appear on live broadcast platforms. In December 2016, Huawei, the Chinese Academy of Sciences, China Academy of Information and Communications Technology, Intel, ARM, etc. jointly initiated the establishment of the Edge Computing Consortium (ECC) in Beijing, aiming to build an edge computing industry cooperation platform, promote open collaboration between operational technology and information and communication technology industries, and promote the healthy and sustainable development of the edge computing industry. Mobile edge computing has the characteristics of low latency, high bandwidth, and location awareness, and is widely used in live broadcast networking, live video optimization, augmented reality and other fields.

[0004] The live broadcast platform also exposed some shortcomings in the process of edge computing application:

[0005] 1: Faced with massive live platform computing tasks and complex heterogeneous edge computing server clusters, it is difficult to design a good task offloading algorithm to improve the offloading success rate of real-time tasks. Task offloading refers to uploading video tasks created by live platform users to the cloud computing center or edge computing server for operation. Depending on the task offloading service provider, it can be divided into local execution, partial offloading, and complete offloading. Different offloading service providers require different offloading strategies, and the complex and heterogeneous edge computing server clusters can easily lead to offloading failures.

[0006] 2: When the resources of the edge computing server are limited, it is difficult to design an efficient CPU resource allocation algorithm for the live broadcast platform, which limits the system's concurrent throughput. The task offloading process of the mobile edge computing of the live broadcast platform can be divided into several stages: task splitting, resource perception, offloading decision, task transmission, execution calculation, and result return. The main factors affecting the system's concurrent throughput include the total amount of resources or resource characteristics of the system itself, the system design, i.e., the organizational structure of the process, and the CPU resource allocation scheme. Among them, the total amount of system resources or resource characteristics are usually fixed and difficult to change. Although many edge computing server clusters can select processor or memory hardware attributes as a reference standard for concurrent throughput, there are still cases where the device model, temperature, instruction set, and even the number of bits of the processor or memory are different, which limits the offloading performance.

[0007] 3: The cost of offloading live edge computing tasks on live platforms is high. The goal of network operators is to maximize revenue by effectively managing and coordinating edge computing service clusters and cloud computing resources, while the goal of live platforms is also to maximize profits. Therefore, the two parties cannot reach an agreement on offloading live edge computing tasks. It is very costly to achieve benign complementarity between network operators and live platforms.

[0008] 4: High energy consumption of live edge computing task offloading on live platforms. Live platforms and network operators provide computing resources and quality of service (QoS) according to their own profit maximization. Computing resources include computing power, storage capacity, power, etc., and QoS includes bandwidth, latency, etc. Therefore, driven by profit maximization, live platforms and network operators rarely care about the energy consumption of live edge computing task offloading, resulting in excessive energy consumption. Summary of the invention

[0009] In view of the above-mentioned problems existing in the prior art, the present invention proposes an intelligent task offloading system for mobile edge computing in a live broadcast platform. The system has a reasonable design and solves the shortcomings of the prior art. It uses a distributed search algorithm to improve the task offloading efficiency while reducing energy consumption and cost requirements.

[0010] An intelligent task offloading system for mobile edge computing in a live broadcast platform, comprising a mobile edge computing device of the live broadcast platform, a base station computing device, and an edge computing server of the live broadcast platform; the mobile edge computing device of the live broadcast platform is connected to the edge computing server of the live broadcast platform through the base station computing device;

[0011] The mobile edge computing device of the live broadcast platform includes a live broadcast edge computing task upload module, a live broadcast edge computing task calculation module, and a plant community intelligent task unloading module;

[0012] Base station computing equipment, including a live edge computing resource perception module, a live edge computing task upload module, a live edge computing task calculation module, a live edge computing task splitting module, a plant community intelligent task unloading module, and a live edge computing result integration module;

[0013] The edge computing server of the live broadcast platform includes a product recommendation computing module, a live broadcast edge computing task computing module, and a plant community intelligent task unloading module;

[0014] The live edge computing task uploading module is used to upload a part of the live edge computing tasks of the live edge computing task computing module to the base station computing device, the live platform edge computing server, and / or other live platform mobile edge computing devices;

[0015] The live broadcast edge computing task calculation module can use the processor and memory in the mobile edge computing device of the live broadcast platform as computing resources and storage resources to realize local computing, and can also unload part of the live broadcast edge computing tasks through the live broadcast edge computing task upload module; the plant community intelligent task unloading module is used to perform computing scheduling and task unloading for the live broadcast edge computing task upload module and the live broadcast edge computing task calculation module in the mobile edge computing device of the live broadcast platform, so as to optimize the utilization efficiency of local computing resources and storage resources; the live broadcast edge computing tasks that are difficult to handle with local computing resources and storage resources are partially unloaded from the live broadcast edge computing task calculation module and uploaded to the base station computing device through the live broadcast edge computing task calculation module, so as to optimize the computing efficiency of the entire live broadcast platform; the plant community intelligent task unloading module on the mobile edge computing device of the live broadcast platform learns from the plant community intelligent task unloading module of the base station computing device and the plant community intelligent task unloading module on the edge computing server of the live broadcast platform, so as to improve the unloading performance from the perspective of the entire live broadcast platform.

[0016] The base station computing device can receive the live edge computing tasks uploaded by the live edge computing task upload module in the live platform mobile edge computing device, perform local processing or further upload to the live platform edge computing server; the base station computing device can receive the calculation results returned by the live platform edge computing server, and integrate them with the local calculation results and then send them to the live platform mobile edge computing device; the live edge computing resource perception module can perceive the computing resources and storage resources in the live platform mobile edge computing device and the live platform edge computing server, and at the same time perceive the live edge computing tasks in the live platform mobile edge computing device and the live platform edge computing server; the live edge computing task upload module is used to unload part of the live edge computing tasks of the live edge computing task calculation module to the live platform edge computing server, or other live platform mobile edge computing devices; the live edge computing task calculation module can use the processor and memory in the base station computing device as computing resources and storage resources to realize local computing, and can also unload part of the live edge computing tasks through the live edge computing task upload module; the live edge computing task splitting module can be based on the live uploaded by the live platform mobile edge computing device The edge computing task is split, a part of the live edge computing task is retained locally to be calculated by the live edge computing task calculation module, and a part of the live edge computing task is uploaded to the live platform edge computing server; the plant community intelligent task unloading module is used to perform calculation scheduling and task unloading for the live edge computing task upload module and the live edge computing task calculation module in the base station computing device, so as to optimize the utilization efficiency of local computing resources and storage resources; for the live edge computing tasks that are difficult to handle with local computing resources and storage resources, a part is unloaded from the live edge computing task calculation module and uploaded to the live platform edge computing server through the live edge computing task upload module, so as to optimize the computing efficiency of the entire live platform; the live edge computing result integration module is used to receive the calculation results of the live platform edge computing server, integrate the calculation results of the local live edge computing task calculation module, and send them to the live platform mobile edge computing device; the plant community intelligent task unloading module of the base station computing device, the plant community intelligent task unloading module on the live platform mobile edge computing device, and the plant community intelligent task unloading module on the live platform edge computing server learn from each other, so as to improve the unloading performance from the perspective of the entire live platform.

[0017] The product recommendation calculation module of the edge computing server of the live broadcast platform is used to solve the product recommendation calculation problem, analyze and calculate the user's viewing behavior, dwell time, interaction frequency and other data on the live broadcast platform, count the user's behavior habits and user needs, and recommend similar videos and related products and services of interest to users; the live broadcast edge computing task calculation module can use the processor and memory in the edge computing server of the live broadcast platform as computing resources and storage resources to realize local computing, and can also unload part of the live broadcast edge computing tasks to other live broadcast platform mobile edge computing devices and live broadcast platform edge computing servers through the live broadcast edge computing task splitting module in the base station computing device; the live broadcast edge computing task calculation module is installed with a virtual machine to shield the differences in unloading tasks on different platforms; the live broadcast edge computing task calculation results of the live broadcast edge computing task calculation module are sent to the live broadcast edge computing result integration module in the base station computing device, and It is further sent to the mobile edge computing device of the live broadcast platform for further processing by the live broadcast edge computing task computing module; the plant community intelligent task unloading module is used to perform calculation scheduling and task unloading on the product recommendation computing module and the live broadcast edge computing task computing module in the edge computing server of the live broadcast platform, so as to optimize the utilization efficiency of local computing resources and storage resources; the live broadcast edge computing tasks that are difficult to handle with local computing resources and storage resources are partially unloaded from the live broadcast edge computing task computing module and unloaded to other live broadcast platform mobile edge computing devices or live broadcast platform edge computing servers through the base station computing device, so as to optimize the computing efficiency of the entire live broadcast platform; the plant community intelligent task unloading module on the edge computing server of the live broadcast platform, the plant community intelligent task unloading module on the mobile edge computing device of the live broadcast platform, and the plant community intelligent task unloading module of the base station computing device can learn from each other and improve the unloading performance from the perspective of the entire live broadcast platform.

[0018] When the system is working, it includes the following steps:

[0019] Step 1: The mobile edge computing device of the live broadcast platform receives the live broadcast edge computing task in the live broadcast edge computing task computing module;

[0020] Step 2: The plant community intelligent task unloading module calls the local resource to calculate the live edge computing task, and uploads the live edge computing task that is difficult to complete to the base station computing device through the live edge computing task uploading module;

[0021] Step 3: The base station computing device receives the live edge computing task uploaded by the live edge computing task uploading module, and senses the live platform mobile edge computing device and live platform edge computing server connected to it through the live edge computing resource sensing module;

[0022] Step 4: The live edge computing task splitting module splits the received live edge computing task under the guidance of the plant community intelligent task unloading module;

[0023] Step 5: The plant community intelligent task offloading module calls local resources and uses the live edge computing task calculation module to calculate the live edge computing task;

[0024] Step 6: The live edge computing task uploading module uploads a part of the live edge computing tasks to the live platform edge computing server;

[0025] Step 7: The live broadcast platform edge computing server receives the live broadcast edge computing task uploaded by the live broadcast edge computing task uploading module;

[0026] Step 8: The plant community intelligent task unloading module calls local resources, uses the live edge computing task calculation module to calculate the live edge computing tasks, and unloads the difficult-to-complete live edge computing tasks to other live platform mobile edge computing devices or live platform edge computing servers through the base station computing device;

[0027] Step 9, the plant community intelligent task offloading module calls local resources, uses the product recommendation calculation module to analyze user preferences, solves the product recommendation calculation problem, and gives the recommendation calculation result;

[0028] Step 10, the calculation result of the live edge computing task of the live edge computing task calculation module and the recommendation calculation result of the product recommendation calculation module are sent to the live edge computing result integration module of the base station computing device;

[0029] Step 11: The base station computing device receives the computing results and recommended computing results of the live edge computing task, integrates them using the live edge computing result integration module, and sends the integrated computing results to the live platform mobile edge computing device;

[0030] Step 12: The mobile edge computing device of the live broadcast platform receives the integrated calculation results, and after the calculation of the live broadcast edge computing task calculation module, returns the calculation results to the user.

[0031] Compared with the prior art, the present invention has the following technical effects:

[0032] 1. The system described in the present invention has a better task offloading success rate; the live broadcast platform mobile edge computing device uses a plant community intelligent task offloading module, and the live broadcast platform edge computing server uses a plant community intelligent task offloading module to solve the task offloading problem of mobile edge computing in the live broadcast platform by simulating the sowing, growth, and fruiting behaviors of the plant community. Faced with a large number of live broadcast platform computing tasks and complex heterogeneous edge computing server clusters, the plant community distributed computing method of this algorithm can easily improve the offloading success rate. After the live broadcast platform mobile edge computing device uploads the task to the live broadcast platform edge computing server, it first queues in the task ready queue and waits to be scheduled for execution. For computing tasks that require complex expertise, such as face recognition, traffic sign detection, etc., they can be deployed in advance on the live broadcast platform edge computing server and then provided to the outside world. When the offloading task source is a heterogeneous architecture, such as the offloading tasks of mobile terminals with different ARM architectures, the live broadcast platform edge computing server installs a virtual machine to shield the differences in offloading tasks on different platforms;

[0033] 2. The system described in the present invention has a higher efficiency in CPU resource allocation; the live broadcast platform mobile edge computing device includes a live broadcast edge computing task calculation module and a plant community intelligent task unloading module; the base station computing device includes a live broadcast edge computing task calculation module and a plant community intelligent task unloading module; the live broadcast platform edge computing server includes a commodity recommendation computing module, a live broadcast edge computing task calculation module, and a plant community intelligent task unloading module; the task splitting, resource perception, unloading decision, task transmission, execution calculation and result feedback of the live broadcast platform mobile edge computing in the task unloading process are allocated to different devices, and the intelligent learning of plant community intelligence is used to optimize the allocation of computing resources and storage resources. Moreover, plant community intelligence is a distributed algorithm. The plant community intelligent task unloading module on the live broadcast platform mobile edge computing device, the plant community intelligent task unloading module of the base station computing device, and the plant community intelligent task unloading module on the live broadcast platform edge computing server can learn from each other, thereby improving the unloading performance;

[0034] 3. The system described in the present invention reduces the cost of offloading live edge computing tasks on the live broadcast platform; the offloading tasks of the mobile edge computing device of the live broadcast platform usually include multiple serial or parallel computing subtasks, and the resources requested by different subtasks vary greatly, including two major categories: CPU-intensive tasks and I / O-intensive tasks. The base station computing device splits the offloading tasks to increase the parallelism of the offloading tasks, and deploys some simple live edge computing tasks locally, thereby improving the execution efficiency of the offloading tasks; with the help of the live edge computing resource perception module of the base station computing device, the status of the connected live broadcast platform mobile edge computing device and the live broadcast platform edge computing server is perceived, and the live edge computing task splitting module splits some task offloading algorithms, and the live broadcast platform edge computing server then makes an offloading decision to understand the service provider information of the live broadcast platform mobile edge computing device after the split, including the computing power, storage capacity, load pressure and network transmission bandwidth available to the live broadcast platform mobile edge computing device, thereby optimizing the offloading decision and reducing the offloading cost;

[0035] 4. The system described in the present invention reduces the energy consumption of live edge computing task unloading of the live broadcast platform. The unloading decision of the live broadcast platform edge computing server is the core link that directly affects the unloading result in the unloading of the live broadcast edge computing task of the live broadcast platform. It outputs whether to unload and the corresponding unloading position, combined with the unloading task requirements and the performance characteristics of the mobile edge computing device of the live broadcast platform. The plant community algorithm can select different optimization objective functions according to different unloading tasks, including task execution delay, unloading energy consumption, etc., to optimize the unloading decision. When the unloading decision result of the mobile edge computing device of the live broadcast platform is local execution, it can be directly skipped. Otherwise, the unloading task data will be uploaded to the base station computing device and the live broadcast platform edge computing server according to complete unloading or partial unloading, such as 4G, 5G cellular network, Wi-Fi. The energy consumption and time consumption of the unloading task transmission of the mobile edge computing device of the live broadcast platform will be calculated in the task unloading target, thereby reducing the network transmission delay and energy consumption during transmission. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0037] Figure 1 It is a flow chart of the working method of the present invention. DETAILED DESCRIPTION

[0038] An intelligent task offloading system for mobile edge computing in a live broadcast platform includes a live broadcast platform mobile edge computing device 100, a base station computing device 200, and a live broadcast platform edge computing server 300; the live broadcast platform mobile edge computing device 100 is connected to the live broadcast platform edge computing server 300 through the base station computing device 200;

[0039] The live broadcast platform mobile edge computing device 100 includes a live broadcast edge computing task uploading module 101, a live broadcast edge computing task calculation module 102, and a plant community intelligent task unloading module 103;

[0040] The base station computing device 200 includes a live edge computing resource perception module 201, a live edge computing task upload module 202, a live edge computing task calculation module 203, a live edge computing task splitting module 204, a plant community intelligent task unloading module 205, and a live edge computing result integration module 206;

[0041] The live broadcast platform edge computing server 300 includes a product recommendation computing module 301, a live broadcast edge computing task computing module 302, and a plant community intelligent task unloading module 303;

[0042] The live broadcast edge computing task uploading module 101 is used to upload a part of the live broadcast edge computing tasks of the live broadcast edge computing task calculation module 102 to the base station computing device 200, the live broadcast platform edge computing server 300, and / or other live broadcast platform mobile edge computing devices 100;

[0043] The live broadcast edge computing task calculation module 102 can use the processor and memory in the mobile edge computing device 100 of the live broadcast platform as computing resources and storage resources to realize local computing, and can also unload part of the live broadcast edge computing tasks through the live broadcast edge computing task upload module 101; the plant community intelligent task unloading module 103 is used to perform computing scheduling and task unloading for the live broadcast edge computing task upload module 101 and the live broadcast edge computing task calculation module 102 in the mobile edge computing device 100 of the live broadcast platform, so as to optimize the utilization efficiency of local computing resources and storage resources; the live broadcast edge computing tasks that are difficult to handle with local computing resources and storage resources are partially unloaded from the live broadcast edge computing task calculation module 102 and uploaded to the base station computing device 200 through the live broadcast edge computing task calculation module 102, so as to optimize the computing efficiency of the entire live broadcast platform; the plant community intelligent task unloading module 103 on the mobile edge computing device 100 of the live broadcast platform learns from each other with the plant community intelligent task unloading module 205 of the base station computing device 200 and the plant community intelligent task unloading module 303 on the live broadcast platform edge computing server 300, so as to improve the unloading performance from the perspective of the entire live broadcast platform.

[0044] The base station computing device 200 can receive the live edge computing tasks uploaded by the live edge computing task upload module 101 in the live platform mobile edge computing device 100, perform local processing or further upload to the live platform edge computing server 300; the base station computing device 200 can receive the calculation results returned by the live platform edge computing server 300, and integrate them with the local calculation results and then send them to the live platform mobile edge computing device 100; the live edge computing resource perception module 201 can perceive the computing resources and storage resources in the live platform mobile edge computing device 100 and the live platform edge computing server 300, and at the same time perceive the live platform mobile edge computing device 100 and the live platform The live edge computing task upload module 202 is used to upload a part of the live edge computing task of the live edge computing task calculation module 203 to the live platform edge computing server 300, or other live platform mobile edge computing devices 100; the live edge computing task calculation module 203 can use the processor and memory in the base station computing device 200 as computing resources and storage resources to realize local computing, and can also unload a part of the live edge computing task through the live edge computing task upload module 202; the live edge computing task splitting module 204 can upload the live edge computing task of the live platform mobile edge computing device 100 according to the live platform mobile edge computing device 100; The live edge computing tasks are split, a part of the live edge computing tasks are retained locally by the live edge computing task calculation module 203, and a part of the live edge computing tasks are uploaded to the live platform edge computing server 300; the plant community intelligent task unloading module 205 is used to perform calculation scheduling and task unloading on the live edge computing task upload module 202 and the live edge computing task calculation module 203 in the base station computing device 200, so as to optimize the utilization efficiency of local computing resources and storage resources; the live edge computing tasks that are difficult to handle with local computing resources and storage resources are unloaded from the live edge computing task calculation module 203 and uploaded through the live edge computing task upload module Block 202 is uploaded to the live broadcast platform edge computing server 300 to optimize the computing efficiency of the entire live broadcast platform; the live broadcast edge computing result integration module 206 is used to receive the computing results of the live broadcast platform edge computing server 300, and integrate the local live broadcast edge computing task computing module 203 computing results, and send them to the live broadcast platform mobile edge computing device 100; the plant community intelligent task unloading module 205 of the base station computing device 200, the plant community intelligent task unloading module 103 on the live broadcast platform mobile edge computing device 100, and the plant community intelligent task unloading module 303 on the live broadcast platform edge computing server 300 learn from each other to improve the unloading performance from the perspective of the entire live broadcast platform.

[0045] The product recommendation calculation module 301 of the live broadcast platform edge computing server 300 is used to solve the product recommendation calculation problem, analyze and calculate the user's viewing behavior, dwell time, interaction frequency and other data on the live broadcast platform, count the user's behavior habits and user needs, and recommend similar videos and related products and services of interest to users; the live broadcast edge computing task calculation module 302 can use the processor and memory in the live broadcast platform edge computing server 300 as computing resources and storage resources to achieve local computing, and can also unload part of the live broadcast edge computing tasks to other live broadcast platform mobile edge computing devices 100 and live broadcast platform edge computing servers 300 through the live broadcast edge computing task splitting module 204 in the base station computing device 200; the live broadcast edge computing task calculation module 302 is installed with a virtual machine to shield the differences in unloading tasks on different platforms; the live broadcast edge computing task calculation results of the live broadcast edge computing task calculation module 302 are sent to the live broadcast edge computing result integration module 206 in the base station computing device 200, and further sent to The live broadcast platform mobile edge computing device 100 is further processed by the live broadcast edge computing task calculation module 102; the plant community intelligent task unloading module 303 is used to perform calculation scheduling and task unloading on the product recommendation calculation module 301 and the live broadcast edge computing task calculation module 302 in the live broadcast platform edge computing server 300, thereby optimizing the utilization efficiency of local computing resources and storage resources; the live broadcast edge computing tasks that are difficult to process with local computing resources and storage resources are partially unloaded from the live broadcast edge computing task calculation module 302 and unloaded to other live broadcast platform mobile edge computing devices 100 or live broadcast platform edge computing servers 300 through the base station computing device 200, thereby optimizing the computing efficiency of the entire live broadcast platform; the plant community intelligent task unloading module 303 on the live broadcast platform edge computing server 300, the plant community intelligent task unloading module 103 on the live broadcast platform mobile edge computing device 100, and the plant community intelligent task unloading module 205 of the base station computing device 200 can learn from each other and improve the unloading performance from the perspective of the entire live broadcast platform.

[0046] When the system is working, it includes the following steps:

[0047] Step 1: The live broadcast platform mobile edge computing device 100 receives the live broadcast edge computing task in the live broadcast edge computing task calculation module 102;

[0048] Step 2, the plant community intelligent task unloading module 103 calls local resources to calculate the live edge computing task, and uploads the live edge computing task that is difficult to complete to the base station computing device 200 through the live edge computing task uploading module 101;

[0049] Step 3: The base station computing device 200 receives the live edge computing task uploaded by the live edge computing task uploading module 101, and senses the live platform mobile edge computing device 100 and the live platform edge computing server 300 connected thereto through the live edge computing resource sensing module 201;

[0050] Step 4, the live edge computing task splitting module 204 splits the received live edge computing task under the guidance of the plant community intelligent task unloading module 205;

[0051] Step 5, the plant community intelligent task unloading module 205 calls local resources and uses the live edge computing task calculation module 203 to calculate the live edge computing task;

[0052] Step 6: The live broadcast edge computing task uploading module 202 uploads a portion of the live broadcast edge computing tasks to the live broadcast platform edge computing server 300;

[0053] Step 7: The live broadcast platform edge computing server 300 receives the live broadcast edge computing task uploaded by the live broadcast edge computing task uploading module 202;

[0054] Step 8: The plant community intelligent task unloading module 303 calls local resources, uses the live edge computing task calculation module 302 to calculate the live edge computing tasks, and unloads the difficult-to-complete live edge computing tasks to other live platform mobile edge computing devices 100 or live platform edge computing servers 300 through the base station computing device 200;

[0055] Step 9, the plant community intelligent task offloading module 303 calls local resources, uses the product recommendation calculation module 301 to analyze user preferences, solves the product recommendation calculation problem, and gives the recommendation calculation result;

[0056] Step 10, the calculation result of the live edge computing task of the live edge computing task calculation module 302 and the recommendation calculation result of the product recommendation calculation module 301 are sent to the live edge computing result integration module 206 of the base station computing device 200;

[0057] Step 11, the base station computing device 200 receives the computing result and the recommended computing result of the live edge computing task, integrates them using the live edge computing result integration module 206, and sends the integrated computing result to the live platform mobile edge computing device 100;

[0058] Step 12: The live broadcast platform mobile edge computing device 100 receives the integrated calculation results, and after the calculation by the live broadcast edge computing task calculation module 102, returns the calculation results to the user.

[0059] Example:

[0060] like Figure 1 As shown in the figure, the intelligent task offloading system of mobile edge computing in the live broadcast platform solves the live broadcast edge computing task offloading problem by simulating the seeding, growth and fruiting behaviors of plant communities.

[0061] An intelligent task offloading system for mobile edge computing in a live broadcast platform includes a live broadcast platform mobile edge computing device 100, a base station computing device 200, and a live broadcast platform edge computing server 300.

[0062] The live broadcast platform mobile edge computing device 100 is a mobile computing device, preferably a smart phone, IPAD, or a portable computer, including a live broadcast edge computing task uploading module 101, a live broadcast edge computing task computing module 102, and a plant community intelligent task unloading module 103; preferably, the CPU core number is 6 cores, the GPU is a six-core graphics processor, stereo dual speakers, and can be used for space perception ultra-wideband chips, with a resolution of 2556x1179px, a pixel density of 460ppi, a screen material OLED, a screen refresh rate of 120Hz, support for HDR technology, a contrast ratio of 2000000:1, a screen brightness of 1000nits typical brightness, 1600nits HDR brightness, and 2000nits peak brightness; the rear camera 1 is the main lens, original color flash, 48 million pixels, aperture f / 1.78, 100% Focus Pixels focus, and second-generation sensor displacement optical image stabilization; the front camera 2 is an ultra-wide-angle lens with a wide angle of 120°, a retina screen flash, an aperture of f / 1.9, and Focus Pixels focus, OIS optical image stabilization, second-generation sensor-shift optical image stabilization; rear video shooting, 4K video shooting, 24fps, 25fps, 30fps or 60fps, 1080p HD video shooting, 25fps, 30fps or 60fps, 720p HD video shooting, 30fps, movie effect mode, up to 4K HDR, 30fps, sports mode, up to 2.8K, 60fps, Dolby Vision HDR video shooting, up to 4K, 60fps; front video shooting, 4K video shooting, 24fps, 25fps, 30fps or 60fps, 1080p HD video shooting, 25fps, 30fps or 60fps, movie effect mode, up to 4K HDR, 30fps, Dolby Vision HDR video shooting, up to 4K, 60fps, ProRes video shooting, up to 4K, 60fps using external storage; network support, 5G network, mobile 5G (NR 5G (NR TDD), China Unicom 5G (NR TDD), China Telecom 5G (NR TDD); 4G network, full network access, China Mobile TD-LTE, China Unicom TD-LTE, China Unicom FDD-LTE, China Telecom TD-LTE, China Telecom FDD-LTE; WLAN function, dual-band WiFi, WiFi6 (IEEE 802.11a / b / g / n / ac / ax), 2x2 MIMO; positioning navigation, Bluetooth 5.3; battery life, video playback up to 23 hours, streaming video playback up to 20 hours, audio playback up to 75 hours.

[0063] The live edge computing task uploading module 101 is a software code, which runs on the live platform mobile edge computing device 100, and can start the network communication module in the live platform mobile edge computing device 100 through system call, and upload the live edge computing task to the base station computing device 200;

[0064] The live edge computing task calculation module 102 is a software code that runs on the live platform mobile edge computing device 100. It can apply for the processor and memory in the live platform mobile edge computing device 100 through system calls, and determine whether the applied processor and memory capacity and duration meet the unloading conditions. If no unloading is required, the live edge computing task is directly executed, and the execution status of the live edge computing task and the processor and memory occupancy are tracked at the same time; if unloading is required, the unfinished live edge computing task is sent to the live edge computing task upload module 101, and unloaded to the live platform edge computing server 300 through the base station computing device 200.

[0065] The plant community intelligent task unloading module 103 is a software code that runs on the live broadcast platform mobile edge computing device 100 and is used to calculate the live broadcast edge computing task unloading problem in the live broadcast platform and search for the optimal feasible solution; the plant community is used to simulate the solution space of the live broadcast edge computing task unloading problem in the live broadcast platform, including multiple plant individuals; further, the plant community intelligent task unloading module 103 on the live broadcast platform mobile edge computing device 100, the plant community intelligent task unloading module 205 of the base station computing device 200, and the plant community intelligent task unloading module 303 on the live broadcast platform edge computing server 300 can learn from each other and improve the unloading performance from the perspective of the entire live broadcast platform.The individual plant is used to simulate a feasible solution to the live edge computing task offloading problem in the live broadcast platform; the plant community population size, that is, the number of plant individuals in the plant community, is used to simulate the number of feasible solutions to the live edge computing task offloading problem in the live broadcast platform; preferably, the number of feasible solutions can be consistent with or inconsistent with the scale of the problem to be solved; the fitness function of the plant, that is, the objective function of the live edge computing task offloading problem in the live broadcast platform, is used to simulate the evaluation of the performance of solving the live edge computing task offloading problem in the live broadcast platform. Users can choose different objective functions as fitness functions according to usage needs and different application scenarios, including offloading delay, bandwidth, reliability, security, Scalability, offloading cost, offloading energy consumption, etc.; the plant community includes seeding operation, growth operation, and result operation; through continuous circulation and iteration of the three operations, the plant community seeks the optimal solution to the problem of offloading the live edge computing task in the live platform; the seeding operation of the plant community simulates the seeding process of the real plant community in nature, which is used to generate the initial feasible solution value required for each round of iterative calculation, and the subsequent growth operation and result operation can be completed using the feasible solution value; the seeding operation is both random and limited to the vicinity of a certain plant plant; the growth operation of the plant community simulates the growth process and mutation process of the real plant community in nature, and only seeds near the water source can continue to grow The individual plants that cannot obtain water sources can only die, and may also mutate due to the influence of water sources or the environment to become new individual plants. The growth operation is used to generate new feasible solutions, which not only retains the main information of the original feasible solutions, but also expands the search range. The growth variation probability of the plant community simulates the probability of mutation of real plants in nature during the growth operation, that is, the feasible solution changes in a small range with the variation probability during the growth process. Furthermore, a larger growth variation probability facilitates the plant community to expand the search space, but also leads to a slower convergence speed. A smaller variation probability is easy to increase the convergence speed of the plant community, but it is also easy to cause the plant community to converge to the local optimal solution earlier. The result operation of the plant community It is used to simulate the learning function of individual plants in real plant communities in nature, and can realize the exchange of part of the information of feasible solutions of different individual plants to generate new feasible solutions, simulating the social learning function of plant communities; the result variation probability of the plant community refers to the probability that different individual plants will generate new feasible solutions through mutual learning and result operation, that is, feasible solutions that are completely different from those of the parent individuals; a smaller result variation probability facilitates the plant community to expand the search space, so that the better solution retains less information, but it will also lead to a slower convergence speed; a larger result variation probability is easy to increase the convergence speed of the plant community, and it is easy to retain more information on the better solutions of the plant community, but it is also easy to cause the plant community to converge to the local optimal solution earlier.

[0066] The base station computing device 200 is the main device of the base station, usually a communication cabinet, which can complete the processing, storage and forwarding of communication data. It includes a live edge computing resource perception module 201, a live edge computing task upload module 202, a live edge computing task calculation module 203, a live edge computing task splitting module 204, a plant community intelligent task unloading module 205, and a live edge computing result integration module 206; preferably, multiple 10 Gigabit Ethernet switches are used, the application layer is layer 2, the processor is MPC8245 or an Intel multi-core processor, the product memory is BOOTROMMB, FLASH MemoryMB or larger, the transmission rate is 10 / 100 / 1000 / 10000Mbps, the switching mode is store-and-forward, the backplane bandwidth and switching capacity are 2.4 / 24Tbps, the packet forwarding rate is 1320Mpps, the port structure is non-modular, the number of ports is 58, and the ports include 24 10 Gigabit SFP+, 24 100M / 1G / 2.5G / 5G / 10G Base-T Ethernet ports, 4 25GE SFP28+2 40GE QSFP+ or 2 100GE QSFP28, 1 expansion slot, supports 2*25GE or 8*10GE optical, 8*25GE photonic card, stackable, supports VxLAN Layer 2 gateway, Layer 3 gateway, supports centralized gateway, distributed gateway, supports BGP-EVPN, supports configuration through Netconf, security management supports encrypted communication analysis (ECA), supports threat trapping technology, and supports full network security defense.

[0067] The live broadcast edge computing resource perception module 201 is a software code that runs on the base station computing device 200. It can obtain information about the live broadcast platform mobile edge computing device 100 and the live broadcast platform edge computing server 300 connected to the base station computing device 200 through system calls, including processor information, memory information, live broadcast edge computing task information and computing resources and storage resource information required to complete the calculation in the live broadcast platform mobile edge computing device 100 and the live broadcast platform edge computing server 300.

[0068] The live edge computing task uploading module 202 is a software code, which runs on the base station computing device 200, and can start the network communication module in the base station computing device 200 through a system call, and upload the live edge computing task to the live platform edge computing server 300;

[0069] The live edge computing task calculation module 203 is a software code that runs on the base station computing device 200. It can apply for the processor and memory in the base station computing device 200 through system calls, and determine whether the capacity and duration of the applied processor and memory meet the unloading conditions. If no unloading is required, the live edge computing task is directly executed, and the execution status of the live edge computing task and the processor and memory occupancy are tracked at the same time; if unloading is required, it will be unloaded to the live platform edge computing server 300 through the live edge computing task upload module 202.

[0070] The live edge computing task splitting module 204 is a software code that runs on the base station computing device 200. It can classify the live edge computing tasks according to the information of the live edge computing tasks and the computing resources and storage resources required to complete the calculation, determine whether it is a professional computing task or whether it needs to be split, and how to split it according to the information perceived by the live edge computing resource perception module 201; further, professional computing tasks require the use of complex professional computing software in the live platform edge computing server 300, including face recognition, traffic sign detection, etc.; further, the live edge computing task splitting module 204 can split a live edge computing task into several serial or parallel computing subtasks, determine the computing resources and storage resources applied for by different subtasks after splitting, and classify the subtasks into two categories: CPU intensive tasks and I / O intensive tasks. Further, CPU intensive tasks apply for more processor resources and usage time, and I / O intensive tasks apply for more I / O operations and memory access.

[0071] The object community intelligent task unloading module 205 is a software code that runs on the base station computing device 200 and is used to calculate the live edge computing task unloading problem in the live broadcast platform and search for the optimal feasible solution; the plant community is used to simulate the solution space of the live edge computing task unloading problem in the live broadcast platform, including multiple plant individuals; further, the plant community intelligent task unloading module 205 of the base station computing device 200, the plant community intelligent task unloading module 103 on the live broadcast platform mobile edge computing device 100, and the plant community intelligent task unloading module 303 on the live broadcast platform edge computing server 300 can learn from each other and improve the unloading performance from the perspective of the entire live broadcast platform.The individual plant is used to simulate a feasible solution to the live edge computing task offloading problem in the live broadcast platform; the plant community population size, that is, the number of individual plant plants in the plant community, is used to simulate the number of feasible solutions to the live edge computing task offloading problem in the live broadcast platform; preferably, the number of feasible solutions can be consistent with or inconsistent with the scale of the problem to be solved; the fitness function of the plant, that is, the objective function of the live edge computing task offloading problem in the live broadcast platform, is used to simulate the evaluation of the performance of solving the live edge computing task offloading problem in the live broadcast platform. Users can choose different objective functions as fitness functions according to usage needs and different application scenarios, including offloading delay, bandwidth, reliability , security, scalability, offloading cost, offloading energy consumption, etc.; the plant community includes seeding operation, growth operation, and result operation; through continuous circulation and iteration of the three operations, the plant community seeks the optimal solution to the problem of offloading the live edge computing task in the live platform; the seeding operation of the plant community simulates the seed sowing process of the real plant community in nature, which is used to generate the initial feasible solution value required for each round of iterative calculation, and the subsequent growth operation and result operation can be completed using the feasible solution value; the seeding operation is both random and limited to the vicinity of a certain plant plant; the growth operation of the plant community simulates the growth process and variation process of the real plant community in nature, and only in the water source Only seeds near the water source can continue to grow, and plant individuals that cannot obtain water sources can only die, and may also mutate due to the influence of water sources or the environment to become new plant individuals; the growth operation is used to generate new feasible solutions, which not only retains the main information of the original feasible solution, but also expands the search range; the growth variation probability of the plant community simulates the probability of mutation of real plants in nature during the growth operation, that is, the feasible solution changes in a small range with the variation probability during the growth process; further, a larger growth variation probability facilitates the plant community to expand the search space, but also leads to a slower convergence speed; a smaller variation probability is easy to increase the convergence speed of the plant community, but it is also easy to cause the plant community to converge to earlier local optimal solution; the result operation of the plant community is used to simulate the learning function of individual plants in real plant communities in nature, which can realize the exchange of part of the information of feasible solutions of different individual plants, generate new feasible solutions, and simulate the social learning function of plant communities; the result variation probability of the plant community refers to the probability that different individual plants learn from each other and perform result operations; a smaller result variation probability facilitates the plant community to expand the search space, so that the better solution retains less information, but it will also lead to a slower convergence speed; a larger result variation probability is easy to increase the convergence speed of the plant community, and it is easy to retain more information on the better solution of the plant community, but it is also easy to cause the plant community to converge to the local optimal solution earlier.

[0072] The live edge computing result integration module 206 is a software code that runs on the base station computing device 200. It can read the splitting of the live edge computing task by the live edge computing task splitting module 204, including information on multiple split sub-tasks, and obtain the calculation results of multiple split sub-tasks of the live edge computing task returned by the live platform edge computing server 300. Preferably, it is the calculation results of multiple split sub-tasks of the live edge computing task returned by multiple live platform edge computing servers 300, and integrate the calculation results of multiple split sub-tasks according to the splitting situation, and transmit the integrated calculation results of the live edge computing task to the live platform mobile edge computing device 100.

[0073] The live broadcast platform edge computing server 300 includes a product recommendation computing module 301, a live broadcast edge computing task computing module 302, and a plant community intelligent task unloading module 303; preferably, it is a server cluster, multiple rack servers, product structure 2U, processor type, IBM POWER8, CPU frequency 3.42GHz, standard CPU quantity 1, L2 cache 512KB, L3 cache 8MB, CPU core ten cores, 1 x8 PCIe slot (must include 4-port 1GB Ethernet LAN available for client use), 6 third-generation concurrent maintenance PCIe slots: 2 x16, 4 third-generation PCIe x8, 2 CAPI adapters per processor module, maximum third-generation PCIe I / O drawer quantity 1, memory type DDR3, memory capacity 64GB, memory description 64GB 1600MHz DDR3, hard disk interface type SAS, standard hard disk capacity 600GB, 2 300GB 15KRPM SAS hot-swappable disk, internal hard disk racks standardly equipped with 12 small SFF or 8 SFF, optional 6 1.8-inch SSD bays, RAID 0, 5, 6, 10; optional: 7200MB cache and Easy Tier function, optical drive DVD (streamlined), dual-port Gigabit network card, PowerVM virtualization software, four-level (L4) cache 16MB / DIMM, processor to memory bandwidth per slot 192GB / s, I / O bandwidth per slot 96GB / s, RAS function real-time partition migration function, processor or memory check error handling, alternative processor recovery, concurrent firmware update, hot-swappable disk bay, hot-swappable concurrent maintenance PCIe slot, hot-swappable and redundant power supply and cooling fan, dynamic processor de-assignment, power supply voltage 100~240V.

[0074] The product recommendation calculation module 301 is a software code that runs on the live broadcast platform edge computing server 300 to provide users with personalized recommendations and product information or suggestions; further, the product recommendation calculation module 301 can analyze the user's live broadcast viewing behavior, interest characteristics and purchasing behavior according to historical data based on the user's live broadcast edge computing tasks received by the live broadcast platform edge computing server 300, and classify users and associate products; further, the product recommendation calculation module 301 can recommend video information and products of interest to users based on the user's information needs, interests, etc., to help users choose which products or services to buy, thereby simulating sales staff to help users complete the purchase process.

[0075] The live broadcast edge computing task calculation module 302 is a software code that runs on the live broadcast platform edge computing server 300. It can apply for the processor and memory in the live broadcast platform edge computing server 300 through system calls, and determine whether the applied processor and memory capacity and duration meet the application requirements of the live broadcast edge computing task and its split subtasks. If no unloading is required, the live broadcast edge computing task and its split subtasks are directly executed, and the execution status and processor and memory occupancy of the live broadcast edge computing task and its split subtasks are tracked at the same time; if further unloading is required, the current live broadcast edge computing task and its split subtasks are transmitted to the base station computing device 200, and further split by the live broadcast edge computing task splitting module 204, and multiple split subtasks are unloaded to other live broadcast platform edge computing servers 300 until the live broadcast edge computing task is completed. Furthermore, the live broadcast platform edge computing server 300 usually uses a heterogeneous edge computing server cluster, some servers have more powerful processor functions, some servers have more powerful I / O functions, and some servers are deployed with professional computing software, such as face recognition, traffic sign detection, etc. The plant community intelligent task offloading module 303 can optimize resource utilization and preferentially offload CPU-intensive tasks, I / O-intensive tasks, and professional computing tasks to the corresponding servers for calculation using the live broadcast edge computing task calculation module 302.

[0076] The plant community intelligent task unloading module 303 is a software code that runs on the live broadcast platform edge computing server 300 and is used to calculate the live broadcast edge computing task unloading problem in the live broadcast platform and search for the optimal feasible solution; the plant community is used to simulate the solution space of the live broadcast edge computing task unloading problem in the live broadcast platform, including multiple plant individuals; further, the plant community intelligent task unloading module 303 on the live broadcast platform edge computing server 300, the plant community intelligent task unloading module 103 on the live broadcast platform mobile edge computing device 100, and the plant community intelligent task unloading module 205 of the base station computing device 200 can learn from each other and improve the unloading performance from the perspective of the entire live broadcast platform.The individual plant is used to simulate a feasible solution to the live edge computing task offloading problem in the live broadcast platform; the plant community population size, that is, the number of individual plant plants in the plant community, is used to simulate the number of feasible solutions to the live edge computing task offloading problem in the live broadcast platform; preferably, the number of feasible solutions can be consistent with or inconsistent with the scale of the problem to be solved; the fitness function of the plant, that is, the objective function of the live edge computing task offloading problem in the live broadcast platform, is used to simulate the evaluation of the performance of solving the live edge computing task offloading problem in the live broadcast platform. Users can choose different objective functions as fitness functions according to usage needs and different application scenarios, including offloading delay, bandwidth, reliability , security, scalability, offloading cost, offloading energy consumption, etc.; the plant community includes seeding operation, growth operation, and result operation; through continuous circulation and iteration of the three operations, the plant community seeks the optimal solution to the problem of offloading the live edge computing task in the live platform; the seeding operation of the plant community simulates the seed sowing process of the real plant community in nature, which is used to generate the initial feasible solution value required for each round of iterative calculation, and the subsequent growth operation and result operation can be completed using the feasible solution value; the seeding operation is both random and limited to the vicinity of a certain plant plant; the growth operation of the plant community simulates the growth process and variation process of the real plant community in nature, and only in the water source Only seeds near the water source can continue to grow, and plant individuals that cannot obtain water sources can only die, and may also mutate due to the influence of water sources or the environment to become new plant individuals; the growth operation is used to generate new feasible solutions, which not only retains the main information of the original feasible solution, but also expands the search range; the growth variation probability of the plant community simulates the probability of mutation of real plants in nature during the growth operation, that is, the feasible solution changes in a small range with the variation probability during the growth process; further, a larger growth variation probability facilitates the plant community to expand the search space, but also leads to a slower convergence speed; a smaller variation probability is easy to increase the convergence speed of the plant community, but it is also easy to cause the plant community to converge to earlier local optimal solution; the result operation of the plant community is used to simulate the learning function of individual plants in real plant communities in nature, which can realize the exchange of part of the information of feasible solutions of different individual plants, generate new feasible solutions, and simulate the social learning function of plant communities; the result variation probability of the plant community refers to the probability that different individual plants learn from each other and perform result operations; a smaller result variation probability facilitates the plant community to expand the search space, so that the better solution retains less information, but it will also lead to a slower convergence speed; a larger result variation probability is easy to increase the convergence speed of the plant community, and it is easy to retain more information on the better solution of the plant community, but it is also easy to cause the plant community to converge to the local optimal solution earlier.

[0077] The system comprises the following steps:

[0078] Step 1: The live streaming platform mobile edge computing device 100 receives a live streaming edge computing task in the live streaming edge computing task calculation module 102;

[0079] Sub-step 1-1: Define the live streaming edge computing task offloading problem as a multi-objective minimization problem; the multi-objective optimization of the minimum value can be understood as a composite function involving multiple objective functions, a series of equality or inequality constraints, and multiple decision variables, and its mathematical expression is:

[0080] Min / Max y = f(x) = (f 1 (x), f 2 (x),..., f i (x)) (i = 1, 2,..., M) (3.1)

[0081]

[0082] In formula (3.1), the objective function is expressed as f(x), where f i (x) is the i-th sub-objective function, and M represents the total number of objective functions to be optimized. In formula (3.2), g(x) and h(x) represent j inequality constraints and k equality constraints respectively, and these constraints together constitute the feasible region of the problem. x = (x 1 , x 2 ,..., x d ) is d decision variables. Based on the above mathematical expression, we can understand that the minimization process in the multi-objective optimization problem aims to solve the decision variable x on the premise of satisfying the constraint conditions g(x) and h(x), so that the values of the m sub-objective functions f m (x) can all reach the optimal value of 0.

[0083] Sub-step 1-2: Define the feasible solution set of the live streaming edge computing task offloading problem; that is, the set of all feasible solutions that satisfy the constraint conditions g(x) and h(x), and the definition expression is:

[0084] A = {x ∈ X | g i (x) ≤ 0; h j (x) = 0} (3.3)

[0085] Sub-step 1-3: Define the Pareto dominance of the live streaming edge computing task offloading problem; let p and q be two decision vectors that satisfy formula (3.3). If p dominates q (p < q), the following two conditions in formula (3.4) need to be satisfied. For all sub-objective functions, the performance of p is not inferior to q, and at the same time, there is at least one sub-objective function such that p is better than q.

[0086]

[0087] Sub-step 1-4, define the Pareto optimal solution and Pareto optimal solution set of the live edge computing task offloading problem; if the current feasible solution x* is not dominated by any solution x in the feasible solution set A, that is, when the constraint conditions of Equation (3.5) are satisfied, then x* is called a Pareto optimal solution in the feasible solution set. The solution set composed of all Pareto optimal solutions that satisfy the constraint conditions is called the Pareto optimal solution set.

[0088]

[0089] Sub-step 1-5, define the Pareto optimal boundary of the live edge computing task offloading problem; the Pareto optimal solution set is evenly distributed on a boundary in the objective function space, and this boundary is called the Pareto optimal boundary. For example, in the objective vector space of a minimization optimization problem with two objectives, f 1 (F) < f 1 (G), f 2 (F) < f 2 (G), which means that the feasible solution F dominates G (F < G), and the feasible solutions A, B, C, D, and E are not dominated by each other. Therefore, the feasible solutions A, B, C, D, and E are Pareto optimal solutions, and the curve evenly distributed among the feasible solutions A, B, C, D, and E is the Pareto optimal boundary.

[0090] PF * = {f(X) = (f 1 (X), f 2 (X),..., f r (X)) | X ∈ {X *}}, {X *} is the optimal solution set (3.6)

[0091] Substeps 1-6 define the intelligent task offloading problem of mobile edge computing in the live broadcast platform with three performance indicators. Considering the relatively large scale of the problem and the relative complexity of the constraints and mathematical models, a multi-objective optimization method based on the Pareto dominance relationship is selected to solve the intelligent task offloading problem of mobile edge computing in the multi-objective live broadcast platform. According to its adaptability, it can effectively handle multi-objective problems and achieve good results in the face of complex constraints and models. The Pareto dominance relationship can be used to find a series of non-dominated solutions that have no obvious superiority or inferiority relationship between multiple objectives, thus providing a more comprehensive selection space for offloading computing decision makers. In the context of multiple objectives, the Pareto dominance method is used to find a set of optimal scheduling solutions to optimize the three performance indicators while taking into account the scale and complexity of the problem. This will help make wise decisions in the intelligent task offloading of mobile edge computing in the live broadcast platform.

[0092] Sub-steps 1-7, the mathematical model and the symbolic definitions of various parameters are as follows;

[0093] n, the total number of live edge computing tasks

[0094] m, total number of processors or memory

[0095] n i , the number of subtasks of the live edge computing task Ji

[0096] O ij , live edge computing tasks J i The j-th subtask, j∈{1, 2, ..., n i}

[0097] M k , the set of all processors or memories, k = 1, 2, ..., m

[0098] M ik , live edge computing tasks J i The set of computable processors or memories for the j-th subtask,

[0099] S ij , live edge computing tasks J i The start calculation time of the j-th subtask

[0100] C ij , live edge computing tasks J i The computation completion time of the j-th subtask

[0101] P ijk , live edge computing tasks J i The j-th subtask O ij In the processor or memory Mk Computation time on

[0102] x ijk , Boolean coefficient; x ijk =1 indicates subtask O ij In the processor or memory M k Calculate on, x ijk =0 indicates subtask O ij Not in processor or memory M k Calculation

[0103] C i , live edge computing tasks J i The actual calculation completion time

[0104] D i , live edge computing tasks J i Estimated time to complete the calculation

[0105] C max , the maximum computation completion time, that is, the time when the last subtask is completed

[0106] Step 2, the plant community intelligent task unloading module 103 calls local resources to calculate the live edge computing task, and uploads the live edge computing task that is difficult to complete to the base station computing device 200 through the live edge computing task uploading module 101;

[0107] The live edge computing task offloading problem is one of the most typical scheduling optimization problems. In the live edge computing task offloading problem, there are multiple live edge computing tasks to be calculated in the edge device. Each live edge computing task has one or more subtasks. These subtasks need to be calculated on a local or remote processor or memory and are subject to some constraints of the subtasks and the processor or memory. The core of the live edge computing task offloading problem is how to reasonably arrange the processor or memory calculation order of the subtasks and determine the start and end time of each subtask on the processor or memory to achieve one or more optimal scheduling goals.

[0108] The intelligent task offloading system of mobile edge computing in the live broadcast platform is an intelligent system for solving the problem of live broadcast edge computing task offloading. Different subtasks of the live broadcast edge computing task can be calculated on multiple processors or memories, and the calculation time of the same subtask on different processors or memories may be different. Therefore, the problem of live broadcast edge computing task offloading needs to consider the reasonable subtask sorting and the calculation of each subtask on the appropriate processor or memory.

[0109] The mathematical definition of the live edge computing task offloading problem includes a set of n live edge computing tasks J i ={J1 , J 2 , ..., J n}, m processors or memory collection M k ={M 1 , M 2 , ..., M m}, each live edge computing task J i Contains i Dao subtask collection There are sequential constraints between subtasks. Each subtask can be offloaded to the local processor or memory collection where the subtask is located. The calculation can be performed on any processor or memory in the system, or offloaded to a processor or memory in a different location, and the calculation time is not fixed.

[0110] Common scheduling performance indicators for the live edge computing task offloading problem cover aspects such as time, cost, and equipment utilization. Time indicators include maximum calculation completion time, maximum lead time, total delay time, etc.; cost indicators involve production costs, maintenance costs, etc.; equipment utilization indicators include total processor or memory load and maximum processor or memory load, etc. Users can choose different indicators according to usage scenarios and personalized needs. Here, only some implementation examples are given, focusing only on the efficiency, user satisfaction, and processor or memory utilization balance of the live broadcast platform. The maximum calculation completion time, total delay time, and total processor or memory load are selected as objective functions to measure the solution performance of the live edge computing task offloading problem. The specific mathematical expression is as follows:

[0111] (1) Minimize the maximum computation completion time

[0112]

[0113] (2) Minimize the total delay time

[0114]

[0115] (3) Minimize the total processor or memory load

[0116]

[0117] (4) The constraints of the live edge computing task offloading problem are:

[0118] s ij +p ijk ·x ijk ≤C ij (3.10)

[0119] C ij ≤s i(j+1) (3.11)

[0120] C i ≤C max (3.12)

[0121] s ij +p ijk ≤s i(j+1) (3.13)

[0122]

[0123] s ij ≥0,C ij ≥0(3.15)

[0124] Formula (3.10) expresses the computation order constraint of the subtasks, that is, the sum of the start computation time and computation time of the live edge computing task cannot exceed the computation completion time of the live edge computing task;

[0125] Formula (3.11) expresses the computation order constraint of subtasks, that is, the computation completion time of the previous subtask of the same live edge computing task must not be later than the start computation time of the next subtask;

[0126] Formula (3.12) expresses the computation completion time constraint of the live edge computing task, ensuring that the computation completion time of each live edge computing task will not exceed the computation completion time of the last live edge computing task;

[0127] Formula (3.13) expresses the computational constraints of the subtasks, indicating that at any time, each processor or memory can only compute one subtask;

[0128] Equation (3.14) expresses the constraints on the computing processor or memory, ensuring that any subtask can only be computed by one processor or memory at any time;

[0129] Formula (3.15) indicates that all time parameters are positive numbers.

[0130] Sub-step 2-1, the plant community intelligent task unloading module 103 initializes the plant community algorithm parameters; the plant community intelligent task unloading module 103 is set including the size of the plant population in the plant community, the numerical type and numerical size of the plant individual, the live edge computing task unloading problem is encoded into a combined optimization problem of the live edge computing task, the processor, and the memory, the objective function of the live edge computing task unloading problem of the plant individual is cleared, the growth probability of the plant community, and the result mutation probability; the processor and memory set of the plant individual are cleared, that is, no processor and memory are selected into the feasible solution, the plant sowing set, the plant community result pairing set, and the plant individual with the highest sorting priority are cleared; the starting and ending conditions of the plant community algorithm are initialized, including the calculation start time, the calculation end time or the iterative calculation number limit, and the end error judgment threshold;

[0131] In sub-step 2-2, the plant community intelligent task unloading module 103 performs a sowing operation; the plant community intelligent task unloading module 103 encodes the plant individuals according to the processor, memory and live edge computing task, and the encoding length depends on the number of processors, memories and live edge computing tasks, and the encoded value depends on whether the processor and memory are selected as live edge computing tasks; preferably, 1 indicates that the processor and memory are selected to serve the live edge computing task, and vice versa, 0 indicates that the processor and memory are not selected to serve the live edge computing task; a plant sowing set is randomly generated according to the plant community population size, and each plant individual in the sowing set is generated as a random number, and the number of plant individuals in the sowing set is the plant community population size;

[0132] In sub-step 2-3, the plant community intelligent task unloading module 103 performs a growth operation on the feasible solution of the live edge computing task unloading problem; the plant community intelligent task unloading module 103 calculates the fitness values ​​of all plant individuals according to the objective function, and compares the fitness values ​​with each other; further, all plant community intelligent task unloading modules 103 select a part of plant individuals with the best fitness to survive according to the growth probability as the feasible solution after the live edge computing task unloading problem grows, and the population size decreases;

[0133] In sub-step 2-4, the plant community intelligent task unloading module 103 performs result operation on the feasible solution of the live edge computing task unloading problem; all surviving plant individuals in the plant community intelligent task unloading module 103 are paired according to the result variation probability, and the two paired plant individuals exchange a part of the binary bits with each other, and generate a part of random solvable according to the result variation probability, that is, the feasible solution of the live edge computing task unloading problem is modified, and the population size increases; further, the plant community intelligent task unloading module 103 uses the feasible solution of the live edge computing task unloading problem as the seed for the next step of sowing;

[0134] In sub-step 2-5, the plant individuals on the plant community intelligent task unloading module 103 are compared and calculated through the fitness function, and it is determined whether the number of iterative calculations meets the predetermined limit on the number of iterative calculations. If so, the plant community intelligent task unloading module 103 ends the calculation, and outputs the plant individual with the highest priority obtained in sub-step 2-5 and the corresponding live edge computing task unloading problem objective function as the optimal solution to the live edge computing task unloading problem; otherwise, if it does not meet the predetermined limit on the number of iterative calculations, the plant community generated by this calculation is used as a seed for re-sowing, and the plant community intelligent task unloading module 103 Return to continue with sub-step 2-2; further, if the local processor and memory in the optimal solution to the live edge computing task offloading problem are sufficient to complete the live edge computing task, there is no need to offload to other edge computing devices, only local offloading is required, and the local processor and memory are scheduled to calculate the live edge computing task according to the optimal solution to the live edge computing task offloading problem; if the local processor and memory in the optimal solution to the live edge computing task offloading problem are insufficient to complete the live edge computing task, it is necessary to offload part of the live edge computing tasks that cannot be completed to other edge computing devices, preferably, to the base station computing device 200;

[0135] Step 3: The base station computing device 200 receives the live edge computing task uploaded by the live edge computing task uploading module 101, and senses the live platform mobile edge computing device 100 and the live platform edge computing server 300 connected thereto through the live edge computing resource sensing module 201;

[0136] Step 4, the live edge computing task splitting module 204 splits the received live edge computing task under the guidance of the plant community intelligent task unloading module 205;

[0137] Step 5, the plant community intelligent task unloading module 205 calls local resources and uses the live edge computing task calculation module 203 to calculate the live edge computing task;

[0138] Sub-step 5-1, the plant community intelligent task unloading module 205 initializes the plant community algorithm parameters; the plant community intelligent task unloading module 205 is set including the size of the plant population in the plant community, the numerical type and numerical size of the plant individual, the live edge computing task unloading problem is encoded into a combined optimization problem of the live edge computing task, the processor, and the memory, the objective function of the live edge computing task unloading problem of the plant individual is cleared, the growth probability of the plant community, and the result mutation probability; the processor and memory set of the plant individual are cleared, that is, no processor and memory are selected into the feasible solution, the plant sowing set, the plant community result pairing set, and the plant individual with the highest sorting priority are cleared; the starting and ending conditions of the plant community algorithm are initialized, including the calculation start time, the calculation end time or the limit of the number of iterative calculations, and the end error judgment threshold;

[0139] In sub-step 5-2, the plant community intelligent task unloading module 205 performs a sowing operation; the plant community intelligent task unloading module 205 encodes the plant individuals according to the processor, memory and live edge computing task, and the encoding length depends on the number of processors, memories and live edge computing tasks, and the encoded value depends on whether the processor and memory are selected as live edge computing tasks; preferably, 1 indicates that the processor and memory are selected to serve the live edge computing task, and vice versa, 0 indicates that the processor and memory are not selected to serve the live edge computing task; a plant sowing set is randomly generated according to the plant community population size, and each plant individual in the sowing set is generated as a random number, and the number of plant individuals in the sowing set is the plant community population size;

[0140] In sub-step 5-3, the plant community intelligent task unloading module 205 performs a growth operation on the feasible solution of the live edge computing task unloading problem; the plant community intelligent task unloading module 205 calculates the fitness values ​​of all plant individuals according to the objective function, and compares the fitness values ​​with each other; further, all plant community intelligent task unloading modules 205 select a part of plant individuals with the best fitness to survive according to the growth probability, as the feasible solution after the growth of the live edge computing task unloading problem, and the population size is reduced;

[0141] Sub-step 5-4, the plant community intelligent task unloading module 205 performs result operation on the feasible solution of the live edge computing task unloading problem; all the surviving plant individuals in the plant community intelligent task unloading module 205 are paired according to the result variation probability, and the two paired plant individuals exchange a part of the binary bits with each other, and generate a part of the random solvable according to the result variation probability, that is, modify the feasible solution of the live edge computing task unloading problem, and increase the population; further, the plant community intelligent task unloading module 205 uses the feasible solution of the live edge computing task unloading problem as the seed for the next step of sowing;

[0142] In sub-step 5-5, the plant individuals on the plant community intelligent task unloading module 205 are compared and calculated through the fitness function, and it is determined whether the number of iterative calculations meets the predetermined limit on the number of iterative calculations. If so, the plant community intelligent task unloading module 205 ends the calculation, and outputs the plant individual with the highest priority obtained in sub-step 5-5 and the corresponding live edge computing task unloading problem objective function as the optimal solution to the live edge computing task unloading problem; otherwise, if it does not meet the predetermined limit on the number of iterative calculations, the plant community generated by this calculation is re-sown as a seed, and the plant community intelligent task unloading module 205 returns Continue with sub-step 5-2; further, if the local processor and memory in the optimal solution to the live edge computing task offloading problem are sufficient to complete the live edge computing task, there is no need to offload to other edge computing devices, only local offloading is required, and the local processor and memory are scheduled to calculate the live edge computing task according to the optimal solution to the live edge computing task offloading problem; if the local processor and memory in the optimal solution to the live edge computing task offloading problem are insufficient to complete the live edge computing task, it is necessary to offload a part of the live edge computing task that cannot be completed to other edge computing devices, preferably, to the live platform edge computing server 300;

[0143] Step 6: The live broadcast edge computing task uploading module 202 uploads a portion of the live broadcast edge computing tasks to the live broadcast platform edge computing server 300;

[0144] Step 7: The live broadcast platform edge computing server 300 receives the live broadcast edge computing task uploaded by the live broadcast edge computing task uploading module 202;

[0145] Step 8: The plant community intelligent task unloading module 303 calls local resources, uses the live edge computing task calculation module 302 to calculate the live edge computing tasks, and unloads the difficult-to-complete live edge computing tasks to other live platform mobile edge computing devices 100 or live platform edge computing servers 300 through the base station computing device 200;

[0146] Sub-step 8-1, the plant community intelligent task unloading module 303 initializes the plant community algorithm parameters; the plant community intelligent task unloading module 303 is set including the size of the plant population in the plant community, the numerical type and numerical size of the plant individual, the live edge computing task unloading problem is encoded into a combined optimization problem of the live edge computing task, the processor, and the memory, the objective function of the live edge computing task unloading problem of the plant individual is cleared, the growth probability of the plant community, and the result mutation probability; the processor and memory set of the plant individual are cleared, that is, no processor and memory are selected into the feasible solution, the plant sowing set, the plant community result pairing set, and the plant individual with the highest sorting priority are cleared; the starting and ending conditions of the plant community algorithm are initialized, including the calculation start time, the calculation end time or the limit of the number of iterative calculations, and the end error judgment threshold;

[0147] In sub-step 8-2, the plant community intelligent task unloading module 303 performs a sowing operation; the plant community intelligent task unloading module 303 encodes the plant individuals according to the processor, memory and live edge computing task, and the encoding length depends on the number of processors, memories and live edge computing tasks, and the encoded value depends on whether the processor and memory are selected as live edge computing tasks; preferably, 1 indicates that the processor and memory are selected to serve the live edge computing task, and vice versa, 0 indicates that the processor and memory are not selected to serve the live edge computing task; a plant sowing set is randomly generated according to the plant community population size, and each plant individual in the sowing set is generated as a random number, and the number of plant individuals in the sowing set is the plant community population size;

[0148] In sub-step 8-3, the plant community intelligent task unloading module 303 performs a growth operation on the feasible solution of the live edge computing task unloading problem; the plant community intelligent task unloading module 303 calculates the fitness values ​​of all plant individuals according to the objective function, and compares the fitness values ​​with each other; further, all plant community intelligent task unloading modules 303 select a part of plant individuals with the best fitness to survive according to the growth probability as the feasible solution after the live edge computing task unloading problem grows, and the population size decreases;

[0149] In sub-step 8-4, the plant community intelligent task unloading module 303 performs result operation on the feasible solution of the live edge computing task unloading problem; all the surviving plant individuals in the plant community intelligent task unloading module 303 are paired according to the result variation probability, and the two paired plant individuals exchange a part of the binary bits with each other, and generate a part of the random solvable according to the result variation probability, that is, the feasible solution of the live edge computing task unloading problem is modified, and the population size increases; further, the plant community intelligent task unloading module 205 uses the feasible solution of the live edge computing task unloading problem as the seed for the next step of sowing;

[0150] In sub-step 8-5, the plant individuals on the plant community intelligent task unloading module 303 are compared and calculated through the fitness function, and it is determined whether the number of iterative calculations meets the predetermined limit on the number of iterative calculations. If so, the plant community intelligent task unloading module 303 ends the calculation, and outputs the plant individual with the highest priority obtained in sub-step 8-5 and the corresponding live edge computing task unloading problem objective function as the optimal solution to the live edge computing task unloading problem; otherwise, if it does not meet the predetermined limit on the number of iterative calculations, the plant community generated by this calculation is re-sown as a seed, and the plant community intelligent task unloading module 303 returns to continue sub-step 8-2 ; Further, if the local processor and memory in the optimal solution to the live edge computing task offloading problem are sufficient to complete the live edge computing task, there is no need to offload to other edge computing devices, only local offloading is required, and the local processor and memory are scheduled to calculate the live edge computing task according to the optimal solution to the live edge computing task offloading problem; if the local processor and memory in the optimal solution to the live edge computing task offloading problem are insufficient to complete the live edge computing task, it is necessary to offload part of the live edge computing tasks that cannot be completed to other edge computing devices, preferably, to other live platform edge computing servers 300 or other live platform mobile edge computing devices 100;

[0151] Step 9, the plant community intelligent task offloading module 303 calls local resources, uses the product recommendation calculation module 301 to analyze user preferences, solves the product recommendation calculation problem, and gives the recommendation calculation result;

[0152] Sub-step 9-1, the plant community intelligent task unloading module 303 initializes the plant community algorithm parameters; the product recommendation calculation module 301 classifies the user preferences according to the historical data of the user live edge computing task, and inputs the user preference classification keywords and the product classification keywords into the plant community intelligent task unloading module 303; further, the plant community intelligent task unloading module 303 is set to include the size of the plant population in the plant community, the numerical type and numerical size of the plant individual, the product recommendation calculation problem is encoded into a combination optimization problem of the user preference classification keywords and the product classification keywords, the objective function of the plant individual product recommendation calculation problem is cleared, the growth probability of the plant community, and the result mutation probability; clear the user preference classification keywords and the product classification keyword set of the plant individual, that is, the user preference classification keywords and the product classification keywords are selected into the feasible solution, clear the plant sowing set, the plant community result pairing set, and clear the plant individual with the highest sorting priority; initialize the starting and ending conditions of the plant community algorithm, including the calculation start time, the calculation end time or the iteration calculation number limit, and the end error judgment threshold;

[0153] In sub-step 9-2, the plant community intelligent task unloading module 303 performs a sowing operation; the plant community intelligent task unloading module 303 encodes the plant plant individuals according to the user preference classification keywords and the commodity classification keywords, the encoding length depends on the number of the user preference classification keywords and the commodity classification keywords, and the encoding value depends on whether the user preference classification keywords and the commodity classification keywords are selected as commodity recommendation calculation tasks; preferably, 1 indicates that the user preference classification keywords and the commodity classification keywords are selected as commodity recommendation calculation tasks, and conversely, 0 indicates that the user preference classification keywords and the commodity classification keywords are not selected as commodity recommendation calculation tasks; a plant plant sowing set is randomly generated according to the plant community population size, each plant plant individual in the sowing set is generated as a random number, and the number of plant plant individuals in the sowing set is the plant community population size;

[0154] Sub-step 9-3, the plant community intelligent task unloading module 303 performs a growth operation on the feasible solution of the commodity recommendation calculation problem; the plant community intelligent task unloading module 303 calculates the fitness values ​​of all plant individuals according to the objective function, and compares the fitness values ​​with each other; further, all plant community intelligent task unloading modules 303 select a part of plant individuals with the best fitness to survive according to the growth probability, as the feasible solution of the commodity recommendation calculation problem after growth, and the population size is reduced;

[0155] Sub-step 9-4, the plant community intelligent task unloading module 303 performs result operation on the feasible solution of the commodity recommendation calculation problem; all the plant plant individuals that survive in the plant community intelligent task unloading module 303 are paired according to the result variation probability, and the two paired plant plant individuals exchange a part of binary bits with each other, and generate a part of random solvable according to the result variation probability, that is, modify the feasible solution of the commodity recommendation calculation problem, and increase the population; further, the plant community intelligent task unloading module 205 uses the feasible solution of the commodity recommendation calculation problem as the seed for the next step of sowing;

[0156] In sub-step 9-5, the plant individuals on the plant community intelligent task unloading module 303 are compared and calculated through the fitness function, and it is determined whether the number of iterative calculations meets the predetermined limit on the number of iterative calculations. If so, the plant community intelligent task unloading module 303 ends the calculation, and outputs the plant individual with the highest priority obtained in sub-step 8-5 and the corresponding commodity recommendation calculation problem objective function as the optimal solution to the commodity recommendation calculation problem; otherwise, if it does not meet the predetermined limit on the number of iterative calculations, the plant community generated by this calculation is re-sown as a seed, and the plant community intelligent task unloading module 303 returns to continue sub-step 8 -2; further, if the local processor and memory in the optimal solution of the product recommendation calculation problem are sufficient to complete the live edge computing task, there is no need to unload to other edge computing devices, only local unloading is required, and the local processor and memory are scheduled to calculate the product recommendation calculation task according to the optimal solution of the product recommendation calculation problem; if the local processor and memory in the optimal solution of the product recommendation calculation problem are insufficient to complete the product recommendation calculation task, it is necessary to unload part of the unfinished product recommendation calculation task to other edge computing devices, preferably, to other live platform edge computing servers 300 or other live platform mobile edge computing devices 100;

[0157] Step 10, the calculation result of the live edge computing task of the live edge computing task calculation module 302 and the recommendation calculation result of the product recommendation calculation module 301 are sent to the live edge computing result integration module 206 of the base station computing device 200;

[0158] Step 11, the base station computing device 200 receives the computing result and the recommended computing result of the live edge computing task, integrates them using the live edge computing result integration module 206, and sends the integrated computing result to the live platform mobile edge computing device 100;

[0159] Step 12: The live broadcast platform mobile edge computing device 100 receives the integrated calculation results, and after the calculation by the live broadcast edge computing task calculation module 102, returns the calculation results to the user.

Claims

1. An intelligent task offloading system for mobile edge computing in a live broadcast platform, characterized in that: It includes a live broadcast platform mobile edge computing device (100), a base station computing device (200), and a live broadcast platform edge computing server (300); the live broadcast platform mobile edge computing device (100) is connected to the live broadcast platform edge computing server (300) via the base station computing device (200); A live broadcast platform mobile edge computing device (100), comprising a live broadcast edge computing task uploading module (101), a live broadcast edge computing task calculation module (102), and a plant community intelligent task unloading module (103); A base station computing device (200), comprising a live edge computing resource perception module (201), a live edge computing task upload module (202), a live edge computing task calculation module (203), a live edge computing task splitting module (204), a plant community intelligent task unloading module (205), and a live edge computing result integration module (206); The live broadcast platform edge computing server (300) includes a product recommendation computing module (301), a live broadcast edge computing task computing module (302), and a plant community intelligent task unloading module (303).

2. The system according to claim 1, characterized in that The live broadcast edge computing task uploading module (101) is used to upload a part of the live broadcast edge computing tasks of the live broadcast edge computing task calculation module (102) to the base station computing device (200), the live broadcast platform edge computing server (300), and / or other live broadcast platform mobile edge computing devices (100); The live edge computing task calculation module (102) can use the processor and memory in the mobile edge computing device (100) of the live platform as computing resources and storage resources to realize local computing, and can also unload a part of the live edge computing tasks through the live edge computing task upload module (101); the plant community intelligent task unloading module (103) is used to perform computing scheduling and task unloading on the live edge computing task upload module (101) and the live edge computing task calculation module (102) in the mobile edge computing device (100) of the live platform, so as to optimize the utilization efficiency of local computing resources and storage resources; local computing resources A portion of the live edge computing tasks that are difficult to process with source and storage resources are unloaded from the live edge computing task calculation module (102) and uploaded to the base station computing device (200) through the live edge computing task calculation module (102), thereby optimizing the computing efficiency of the entire live platform; the plant community intelligent task unloading module (103) on the live platform mobile edge computing device (100) learns from each other with the plant community intelligent task unloading module (205) of the base station computing device (200) and the plant community intelligent task unloading module (303) on the live platform edge computing server (300), thereby improving the unloading performance from the perspective of the entire live platform.

3. The system according to claim 1, characterized in that The base station computing device (200) is capable of receiving the live broadcast edge computing task uploaded by the live broadcast edge computing task uploading module (101) in the live broadcast platform mobile edge computing device (100), and performing local processing or further uploading to the live broadcast platform edge computing server (300); the base station computing device (200) is capable of receiving the computing result returned by the live broadcast platform edge computing server (300), and integrating it with the local computing result and then sending it to the live broadcast platform mobile edge computing device (100); The live broadcast edge computing resource perception module (201) is capable of perceiving computing resources and storage resources in the live broadcast platform mobile edge computing device (100) and the live broadcast platform edge computing server (300), and simultaneously perceiving the live broadcast edge computing tasks in the live broadcast platform mobile edge computing device (100) and the live broadcast platform edge computing server (300); The live broadcast edge computing task uploading module (202) is used to upload a part of the live broadcast edge computing task of the live broadcast edge computing task computing module (203) to the live broadcast platform edge computing server (300), or other live broadcast platform mobile edge computing devices (100); the live broadcast edge computing task computing module (203) can use the processor and memory in the base station computing device (200) as computing resources and storage resources to realize local computing, and can also unload a part of the live broadcast edge computing task through the live broadcast edge computing task uploading module (202); The live edge computing task splitting module (204) can split the live edge computing tasks uploaded by the live platform mobile edge computing device (100), retain a part of the live edge computing tasks locally to be calculated by the live edge computing task calculation module (203), and upload a part of the live edge computing tasks to the live platform edge computing server (300); the plant community intelligent task unloading module (205) is used to perform calculation scheduling and task unloading on the live edge computing task upload module (202) and the live edge computing task calculation module (203) in the base station computing device (200), thereby optimizing the utilization efficiency of local computing resources and storage resources; for the live edge computing tasks that are difficult to process with local computing resources and storage resources, a part is unloaded from the live edge computing task calculation module (203) and uploaded to the live platform edge computing server (300) through the live edge computing task upload module (202), thereby optimizing the computing efficiency of the entire live platform; The live broadcast edge computing result integration module (206) is used to receive the computing results of the live broadcast platform edge computing server (300), and integrate the computing results of the local live broadcast edge computing task computing module (203), and send them to the live broadcast platform mobile edge computing device (100); the plant community intelligent task unloading module (205) of the base station computing device (200), the plant community intelligent task unloading module (103) on the live broadcast platform mobile edge computing device (100), and the plant community intelligent task unloading module (303) on the live broadcast platform edge computing server (300) learn from each other to improve the unloading performance from the perspective of the entire live broadcast platform.

4. The system according to claim 1, characterized in that The commodity recommendation calculation module (301) of the live broadcast platform edge computing server (300) is used to solve the commodity recommendation calculation problem, analyze and calculate the user's viewing behavior, residence time, interaction frequency and other data on the live broadcast platform, count the user's behavior habits and user needs, and recommend similar videos and related products and services of interest to the user; the live broadcast edge computing task calculation module (302) can use the processor and memory in the live broadcast platform edge computing server (300) as computing resources and storage resources to realize local computing, and can also unload a part of the live broadcast edge computing tasks to other live broadcast platform mobile edge computing devices (100) and live broadcast platform edge computing servers (300) through the live broadcast edge computing task splitting module (204) in the base station computing device (200); The live edge computing task calculation module (302) is installed with a virtual machine to shield the differences in unloading tasks on different platforms; The result of the live edge computing task calculation of the live edge computing task calculation module (302) is sent to the live edge computing result integration module (206) in the base station computing device (200), and further sent to the live platform mobile edge computing device (100) for further processing by the live edge computing task calculation module (102); the plant community intelligent task unloading module (303) is used to perform calculation scheduling and task unloading on the commodity recommendation calculation module (301) and the live edge computing task calculation module (302) in the live platform edge computing server (300), thereby optimizing the utilization efficiency of local computing resources and storage resources; the live edge computing tasks that are difficult to process with local computing resources and storage resources are partially unloaded from the live edge computing task calculation module (302) and unloaded to other live platform mobile edge computing devices (100) or live platform edge computing servers (300) through the base station computing device (200), thereby optimizing the computing efficiency of the entire live platform; The plant community intelligent task unloading module (303) on the live broadcast platform edge computing server (300), the plant community intelligent task unloading module (103) on the live broadcast platform mobile edge computing device (100), and the plant community intelligent task unloading module (205) of the base station computing device (200) can learn from each other and improve the unloading performance from the perspective of the entire live broadcast platform.

5. The system according to claim 1, characterized in that When the system is working, it includes the following steps: Step 1: The live broadcast platform mobile edge computing device (100) receives the live broadcast edge computing task in the live broadcast edge computing task calculation module (102); Step 2, the plant community intelligent task unloading module (103) calls local resources to calculate the live edge computing task, and uploads the live edge computing task that is difficult to complete to the base station computing device (200) through the live edge computing task uploading module (101); Step 3, the base station computing device (200) receives the live broadcast edge computing task uploaded by the live broadcast edge computing task uploading module (101), and senses the live broadcast platform mobile edge computing device (100) and the live broadcast platform edge computing server (300) connected thereto through the live broadcast edge computing resource sensing module (201); Step 4, the live edge computing task splitting module (204) splits the received live edge computing task under the guidance of the plant community intelligent task unloading module (205); Step 5, the plant community intelligent task unloading module (205) calls local resources and uses the live edge computing task calculation module (203) to calculate the live edge computing task; Step 6, the live broadcast edge computing task uploading module (202) uploads a portion of the live broadcast edge computing tasks to the live broadcast platform edge computing server (300); Step 7, the live broadcast platform edge computing server (300) receives the live broadcast edge computing task uploaded by the live broadcast edge computing task uploading module (202); Step 8, the plant community intelligent task unloading module (303) calls local resources, uses the live edge computing task calculation module (302) to calculate the live edge computing tasks, and unloads the live edge computing tasks that are difficult to complete to other live platform mobile edge computing devices (100) or live platform edge computing servers (300) through the base station computing device (200); Step 9, the plant community intelligent task unloading module (303) calls local resources, uses the product recommendation calculation module (301) to analyze user preferences, solves the product recommendation calculation problem, and gives the recommendation calculation result; Step 10, the calculation result of the live edge computing task of the live edge computing task calculation module (302) and the recommendation calculation result of the product recommendation calculation module (301) are sent to the live edge computing result integration module (206) of the base station computing device (200); Step 11, the base station computing device (200) receives the computing result and the recommended computing result of the live broadcast edge computing task, integrates them using the live broadcast edge computing result integration module (206), and sends the integrated computing result to the live broadcast platform mobile edge computing device (100); Step 12, the live broadcast platform mobile edge computing device (100) receives the integrated calculation results, and after the calculation by the live broadcast edge computing task calculation module (102), returns the calculation results to the user.

Images