Multi-level large-scale AI computing power intelligent scheduling system and scheduling method

Abstract

The invention discloses a multi-level large-scale AI computing power intelligent scheduling system and scheduling method. The system comprises an AI algorithm and alarm management platform at a group side and an AI algorithm service platform at an enterprise side. The AI algorithm service platform is used for receiving the monitoring video, generating alarm information through analysis and uploading the alarm information to the AI algorithm and alarm management platform; and the AI algorithm and alarm management platform is used for updating an AI algorithm arrangement task according to the alarm information and issuing the AI algorithm arrangement task to the AI algorithm service platform. According to the method, AI computing power resources of all enterprises in the group can be optimally allocated and utilized, so that the resource utilization rate is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of AI computing power intelligent scheduling, and more specifically to a multi-level large-scale AI computing power intelligent scheduling system and scheduling method. Background Art

[0002] In today's wave of industrial automation and informatization, surveillance video systems have become an indispensable part of corporate production sites. By installing surveillance cameras, companies can monitor the production process in real time, and promptly detect and deal with illegal operations, process deviations, and product quality issues. With the development of artificial intelligence (AI) technology, AI intelligent analysis algorithms are widely used in surveillance video data to improve monitoring efficiency and intelligence.

[0003] At present, large group companies with many subsidiaries face the following problems when using AI intelligent analysis algorithms to analyze and process production site surveillance videos:

[0004] There are many subordinate enterprises, and the amount of surveillance video data at each production site is huge, so the number of video streams that need to be processed is huge; different production scenarios and monitoring requirements require different AI algorithms, resulting in a large number of AI algorithms that need to be run;

[0005] However, when conducting 24-hour real-time AI analysis on all surveillance videos, the demand for computing resources is extremely high, and some companies cannot afford such a huge computing cost;

[0006] Moreover, the management levels, safety risks, and probabilities of quality problems vary greatly among different companies, resulting in unreasonable allocation of AI computing resources and the contradiction between resource waste and resource shortage.

[0007] From a group perspective, the AI ​​computing resources of various companies have not been optimally allocated and utilized, affecting overall production efficiency and management effectiveness.

[0008] In response to the above problems, this application proposes a multi-level large-scale AI computing power intelligent scheduling system. Summary of the invention

[0009] The multi-level large-scale AI computing power intelligent scheduling system and scheduling method disclosed in this application are intended to solve the computing power resource allocation problem faced by large group-type enterprises in AI surveillance video analysis, and to achieve optimal utilization of computing power resources through intelligent scheduling methods, thereby improving the production management level of the entire group and the efficiency of AI computing power utilization.

[0010] In order to achieve the above object, the present invention adopts the following technical solution:

[0011] A multi-level large-scale AI computing power intelligent scheduling system, including an AI algorithm and alarm management platform on the group side and an AI algorithm service platform on the enterprise side;

[0012] AI algorithm service platform, which is used to receive surveillance videos, generate alarm information through analysis, and upload the alarm information to the AI ​​algorithm and alarm management platform;

[0013] The AI ​​algorithm and alarm management platform is used to update the AI ​​algorithm orchestration tasks based on the alarm information and send them to the AI ​​algorithm service platform.

[0014] Preferably, the AI ​​algorithm and alarm management platform includes an alarm receiving module, an alarm intelligent analysis module, and a computing power intelligent management and scheduling module;

[0015] An alarm receiving module is used to receive alarm information and send it to the alarm intelligent analysis module;

[0016] The alarm intelligent analysis module is used to analyze the alarm information to obtain the alarm frequency of the surveillance videos of all production sites of each enterprise, and the alarm frequency of each production site surveillance video in each period;

[0017] The intelligent computing power management and scheduling module is used to update the AI ​​algorithm to schedule and distribute tasks based on the frequency and computing power resource information of each enterprise.

[0018] Update AI algorithm orchestration tasks include:

[0019] Perform 24-hour real-time AI algorithm analysis and processing on surveillance videos of high-risk production sites of enterprises;

[0020] Conduct intelligent video inspections on other surveillance videos of the enterprise;

[0021] And evaluate the enterprise's computing resources and generate scheduling instructions based on the evaluation results.

[0022] Preferably, the AI ​​algorithm service platform includes an AI analysis module and a video receiving module;

[0023] Video receiving module, used to receive surveillance video data and send it to the AI ​​analysis module;

[0024] The AI ​​analysis module is used to receive and analyze the monitoring video data according to the AI ​​algorithm orchestration tasks issued by the computing power intelligent management and scheduling module to generate alarm information, and send the alarm information to the alarm receiving module.

[0025] The alarm information includes: enterprise coding information, video source coding information, alarm type, computing resource information, alarm generation time, image screenshots and / or video clips that generated the alarm.

[0026] Preferably, the AI ​​algorithm and alarm management platform further includes a monitoring image receiving module 1 and a monitoring image sending module;

[0027] Monitoring image receiving module 1, used to receive monitoring videos to be processed in enterprises with insufficient computing power;

[0028] The monitoring image sending module is used to receive and send the surveillance videos to be processed to enterprises with sufficient computing power according to the AI ​​algorithm scheduling tasks.

[0029] Preferably, the AI ​​algorithm and alarm management platform also includes a monitoring image storage module, which is used to store the monitoring video to be processed received by the monitoring image receiving module 1, and send the monitoring video to be processed to the monitoring image sending module.

[0030] Preferably, the AI ​​algorithm service platform further includes a monitoring image interception and transmission module and a monitoring image receiving module 1;

[0031] A monitoring image interception and transmission module is used to receive the monitoring video sent by the video receiving module and upload it to the monitoring image receiving module 1;

[0032] The monitoring image receiving module 2 is used to receive the monitoring video to be processed sent by the monitoring image sending module and send it to the AI ​​analysis module.

[0033] Preferably, the computing power intelligent management and scheduling module is configured with an information reporting service for collecting the computing power information of the enterprise for initialization; the initialization steps include:

[0034] Obtain computing power information of each enterprise, including computing resource information, surveillance video information, and required AI algorithm information;

[0035] Allocate weight values ​​according to the computing power information of each enterprise, including allocating initial weight values ​​to the surveillance videos of each enterprise and allocating initial weight values ​​to the time period of each surveillance video;

[0036] Generate initial algorithm scheduling tasks based on weight values ​​and computing resource information.

[0037] Preferably, the goal of allocating weight values ​​is to maximize the AI ​​computing resource utilization score.

[0038] In addition, the present application also provides a multi-level large-scale AI computing power intelligent scheduling method, which applies the multi-level large-scale AI computing power intelligent scheduling system as described above, including:

[0039] The enterprise side receives the surveillance video, generates alarm information through analysis, and uploads the alarm information to the group side;

[0040] The group updates the AI ​​algorithm to schedule tasks based on the alarm information and sends them to the enterprise.

[0041] Preferably, updating the AI ​​algorithm orchestration task includes classifying the computing power of enterprises, including insufficient computing power and sufficient computing power; and collecting videos uploaded by enterprises with insufficient computing power and sending them to enterprises with sufficient computing power.

[0042] Through the above technical solutions, it can be known that the present invention discloses a multi-level large-scale AI computing power intelligent scheduling system and scheduling method, which can perform differentiated processing on enterprise monitoring videos and intelligently adjust AI algorithm scheduling tasks according to the alarm frequency and computing power resource information of enterprise monitoring videos, effectively reduce the enterprise's investment in AI computing power resources and alleviate the burden on the enterprise; at the same time, it provides customized AI algorithm analysis for the production sites of different enterprises, which not only realizes resource sharing within the group, but also enables the AI ​​computing power resources of each enterprise within the group to be optimally allocated and utilized, thereby improving resource utilization and enhancing the overall production efficiency and management effectiveness of the group.

[0043] Compared with the existing technology, the present invention can efficiently process massive surveillance video data from many subordinate enterprises, solve the processing problems caused by the huge amount of data, and ensure the real-time and accuracy of video analysis; it also realizes the reasonable allocation and optimal use of AI computing resources, effectively alleviates the problem of tight computing resources, avoids the contradiction between resource waste and shortage, and reduces the operating costs of enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0045] Figure 1 This is a schematic diagram of the overall architecture of a multi-level large-scale AI computing power intelligent scheduling system.

[0046] Figure 2 This is a flowchart of the AI ​​computing power intelligent scheduling system scheduling computing power for enterprises with insufficient computing power.

[0047] Figure 3 This is a flow chart of the AI ​​computing power intelligent scheduling system scheduling computing power for enterprises with abundant computing power.

[0048] Figure 4 This is a schematic diagram of the initialization process for the first run of the AI ​​computing power intelligent scheduling system. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] The present invention involves advanced technologies such as big data analysis, artificial intelligence, and cloud-edge collaboration. Its purpose is to maximize the use of the AI ​​computing resources of each enterprise at the group level through video intelligent inspection and AI computing power intelligent scheduling methods when analyzing and processing enterprise surveillance videos through AI analysis algorithms for large group-type enterprises with a large number of subordinate enterprises, thereby improving the overall efficiency of AI computing power utilization and realizing intelligent scheduling of edge-side AI computing resources.

[0051] Embodiment 1:

[0052] The embodiment of the present invention discloses a multi-level large-scale AI computing power intelligent scheduling system, including an AI algorithm and alarm management platform on the group side and an AI algorithm service platform on the enterprise side;

[0053] AI algorithm service platform, which is used to receive surveillance videos, generate alarm information through analysis, and upload the alarm information to the AI ​​algorithm and alarm management platform;

[0054] The AI ​​algorithm and alarm management platform is used to update the AI ​​algorithm orchestration tasks based on the alarm information and send them to the AI ​​algorithm service platform.

[0055] Reference Figure 1 , Figure 1 This is a structural diagram of the multi-level large-scale AI computing power intelligent scheduling system of this embodiment; wherein,

[0056] The AI ​​algorithm and alarm management platform 7 can receive the alarm information after AI analysis and processing reported by the AI ​​algorithm service platforms of each enterprise under the group. After analyzing and processing these alarm information, it intelligently adjusts the algorithm tasks of each enterprise's AI algorithm service platform and sends the algorithm task scheduling instructions to each enterprise's AI algorithm service platform.

[0057] The AI ​​algorithm and alarm management platform 7 specifically includes an alarm receiving module 13, an alarm intelligent analysis module 10, and a computing power intelligent management and scheduling module 11;

[0058] The alarm receiving module 13 is used to receive the alarm information and send it to the alarm intelligent analysis module;

[0059] The alarm intelligent analysis module 10 is used to analyze the alarm information to obtain the alarm frequency of the monitoring videos of all production sites of each enterprise and the alarm frequency of each production site monitoring video in each period;

[0060] The computing power intelligent management and scheduling module 11 is used to update the AI ​​algorithm scheduling tasks and issue them according to the frequency and computing power resource information of each enterprise.

[0061] Update AI algorithm orchestration tasks include:

[0062] Perform 24-hour real-time AI algorithm analysis and processing on surveillance videos of high-risk production sites of enterprises;

[0063] Conduct intelligent video inspections on other surveillance videos of the enterprise;

[0064] That is, AI analysis and processing are performed on surveillance videos in different time periods, and more AI computing power is allocated to time periods where risks are likely to occur.

[0065] In one embodiment, it also includes evaluating the enterprise computing resources and generating scheduling instructions based on the evaluation results.

[0066] In this embodiment, the AI ​​algorithm and alarm management platform 7 can obtain the AI ​​computing power resource status of each subordinate enterprise, and combine the alarm analysis results and algorithm task requirements of each enterprise to determine whether each enterprise has surplus computing power resources.

[0067] For enterprises with insufficient AI computing resources, the group platform will issue instructions to the enterprise's AI algorithm service platform to upload video surveillance images that do not have excess computing resources for analysis and processing to the group.

[0068] For enterprises with abundant computing resources, the group platform will issue instructions to the enterprise's AI algorithm service platform to receive the previous monitoring images and corresponding algorithms of other enterprises with insufficient computing resources from the group platform, run the algorithm to process the monitoring images of other enterprises, and support other enterprises in conducting AI algorithm analysis.

[0069] In an exemplary embodiment, the AI ​​algorithm and alarm management platform 7 further includes a monitoring image receiving module 8 and a monitoring image sending module 12;

[0070] Monitoring image receiving module 1, used to receive monitoring videos to be processed in enterprises with insufficient computing power;

[0071] The monitoring image sending module is used to receive and send the surveillance videos to be processed to enterprises with sufficient computing power according to the AI ​​algorithm scheduling tasks.

[0072] Preferably, the AI ​​algorithm and alarm management platform also includes a monitoring image storage module 9, which is used to store the monitoring video to be processed received by the monitoring image receiving module 8, and send the video to be processed to the monitoring image sending module 12.

[0073] In one embodiment, if Figure 1 , taking enterprise A as an example;

[0074] The AI ​​algorithm service platform on the enterprise side includes an AI analysis module 5 and a video receiving module 3;

[0075] The video receiving module 3 is used to receive the monitoring video data and send it to the AI ​​analysis module. In this embodiment, the video receiving module 3 is connected to the enterprise video source 1 to receive the monitoring video data in the video source 1. At the same time, according to the video transmission protocol of the enterprise video source (1), the video data is parsed and extracted.

[0076] The AI ​​analysis module 5 is used to receive and analyze the monitoring video data according to the AI ​​algorithm scheduling task issued by the computing power intelligent management and scheduling module 11 to generate alarm information, and send the alarm information to the alarm receiving module 13.

[0077] The alarm information includes: enterprise coding information, video source coding information, alarm type, computing resource information, alarm generation time, image screenshots and / or video clips that generated the alarm, and other information.

[0078] Further, in order to assist the group in completing computing power scheduling, the AI ​​algorithm service platform 2 also includes a monitoring image interception and transmission module 4 and a monitoring image receiving module 2 6;

[0079] The monitoring image interception and transmission module 4 is used to receive the monitoring video sent by the video receiving module 3 and upload it to the monitoring image receiving module 1 8;

[0080] The monitoring image receiving module 2 6 is used to receive the monitoring video to be processed sent by the monitoring image sending module 12 and send it to the AI ​​analysis module 5.

[0081] In this embodiment, the working process of the multi-level large-scale AI computing power intelligent scheduling system during operation is as follows:

[0082] 1) The video receiving module 2 is used to receive the monitoring video data from the enterprise video source 1, and according to the video transmission protocol of the enterprise video source 1, parse and extract the video data and send it to the AI ​​analysis module 5; the AI ​​analysis module 5 performs 24-hour real-time AI algorithm analysis and processing on the specified monitoring video in the video received from the video receiving module 3, or AI algorithm analysis in different time periods, i.e., intelligent inspection, according to the algorithm scheduling tasks received from the computing power intelligent management and scheduling module 11; and uploads the alarm information found in the analysis to the alarm receiving module 13;

[0083] The alarm receiving module 13 sends the received alarm information to the alarm intelligent analysis module 10, which is used to analyze the frequency of alarms in the surveillance videos of all production sites of each enterprise, as well as the frequency of alarms in each time period of the surveillance videos of each production site, and send the analysis results to the computing power intelligent management and scheduling module 11; the computing power intelligent management and scheduling module 11, based on the analysis results given by the alarm intelligent analysis module, combined with the computing power resource situation of each enterprise and the weight parameters of different risks, through intelligent analysis, finally determines which surveillance videos and algorithms of each enterprise need to run in real time 24 hours a day, and which surveillance videos can be inspected in different time periods, and sends these algorithm scheduling tasks to the AI ​​analysis module in the AI ​​algorithm service platform of the corresponding enterprise, and then the AI ​​analysis module 5 analyzes the surveillance video data according to the newly issued AI algorithm scheduling tasks.

[0084] 2) If the computing power intelligent management and scheduling module 11 analyzes that the computing power resources of some enterprises are insufficient, the image capture and sending module 4 receives the specified video from the video receiving module 3 according to the instructions from the computing power intelligent management and scheduling module 11, and first decodes the received monitoring video, and then extracts a fixed number of image frames per second from the decoded YUV image and encodes it into JPEG and other types of images; finally, the captured image is sent to the monitoring image receiving module 8, and the monitoring image receiving module 8 stores the monitoring image in the monitoring image storage module 9.

[0085] When it is analyzed that some enterprises have abundant computing resources, the monitoring image sending module 12 reads the video data from the monitoring image storage module 9 according to the instructions of the computing power intelligent management and scheduling module 11, and sends it to the monitoring image receiving module 2 6 through the monitoring image sending module 12; after the monitoring image receiving module 2 6 receives the monitoring image sent by the monitoring image sending module 12, it is further sent to the AI ​​analysis module 5 for analysis and processing. The AI ​​analysis module 5 finally uploads the alarm information discovered by the analysis to the alarm receiving module 13 on the group side. At this time, the alarm information includes the source information of the monitoring image.

[0086] In an exemplary embodiment, referring to Figure 2 , and taking the insufficient computing power of enterprise A as an example, the AI ​​computing power scheduling steps include:

[0087] Step 1: The computing power intelligent management and scheduling module 11 in the group-side AI algorithm and alarm management platform 7 receives the probability of the monitored problem appearing in each monitoring video of each enterprise from the alarm intelligent analysis module 10, including the probability of the monitored problem appearing in each time period of the monitoring video. In a certain period, such as one week or one month, the algorithm scheduling task for each enterprise is recalculated based on the monitored problem information and the weight information of the monitoring video, combined with the computing power resource information of all the enterprises under the group.

[0088] Orchestration tasks include:

[0089] 1) According to the weight value of each surveillance video, the weight value of each time period in the surveillance video, the enterprise computing power resource status and other data, calculate the AI ​​computing resource utilization score of each enterprise's algorithm orchestration task, and at the same time calculate the total AI computing resource utilization score of all enterprise algorithm orchestration tasks under the group. After repeated iterations, the highest total AI computing resource utilization score of all enterprise algorithm orchestration tasks under the group is selected to determine the final result of all enterprise algorithm orchestration tasks in this round of adjustment.

[0090] 2) Based on whether the enterprise's AI computing power resources meet the requirements of algorithm orchestration tasks, the computing power resources of each enterprise are classified into three categories: sufficient computing power, appropriate computing power, and insufficient computing power.

[0091] For enterprises with insufficient computing power, appropriate computing resources are analyzed and determined from enterprises with sufficient computing power, and in the algorithm orchestration tasks of the enterprise with insufficient computing power, the enterprise platform is required to upload image screenshots of surveillance videos that cannot be processed by the local algorithm service platform to the group platform; for enterprises with sufficient computing power, in the algorithm orchestration tasks of the enterprise, the enterprise platform is required to receive screenshots of surveillance videos of designated enterprises from the group platform.

[0092] Step 2: The computing power intelligent management and scheduling module 11 sends the task scheduling instructions of enterprise A to the enterprise-side AI algorithm service platform 2. The sent task scheduling instructions also require enterprise A to upload the image group of the specified surveillance video to the group side.

[0093] Step 3: The AI ​​analysis module 5 in the enterprise-side AI algorithm service platform 2 saves the received algorithm orchestration tasks and algorithm application files in the enterprise's local AI algorithm service platform.

[0094] It includes analyzing the algorithm arrangement tasks received by the enterprise, notifying the video receiving module 3 according to the algorithm arrangement task information, pulling the monitoring video from the enterprise video source 1, and pulling the video stream 24 hours a day for the video that needs to run the algorithm in real time 24 hours a day; for the video inspection that needs to be carried out in time periods, pull the corresponding video stream during the required time period. And according to the algorithm arrangement task, load the corresponding AI algorithm for analysis and processing of the received monitoring video.

[0095] And the problems found when executing the AI ​​analysis algorithm generate alarm information and send it to the alarm receiving module 13 on the group side. The subsequent execution steps of the alarm receiving module 13 on the group side, the alarm intelligent analysis module 10, and the computing power intelligent management and scheduling module 11 are the same as described above.

[0096] Step 4: The monitoring image capture and sending module 4 simultaneously arranges tasks according to the algorithm, receives the specified monitoring video from the video receiving module 3, decodes the video, captures the video screenshots of the specified time period and saves them as JPEG images, and then uploads the images to the group-side monitoring image receiving module 8; the monitoring image receiving module 8 further sends the images to the monitoring image storage module 9.

[0097] Further references Figure 3 This application enables enterprises with abundant computing power to output computing power to support enterprises with insufficient computing power. In this embodiment, the example of enterprise A with abundant computing power is used to illustrate again. The AI ​​computing power scheduling steps include:

[0098] Step 1: Same as step 1 in the previous case of insufficient computing power;

[0099] Step 2: The computing power intelligent management and scheduling module 11 on the group side sends the algorithm orchestration task of enterprise A to the AI ​​algorithm service platform 2 on the enterprise side. The sent algorithm orchestration task also requires enterprise A to receive monitoring images from the group side.

[0100] That is, the AI ​​analysis module 5 saves the received algorithm arrangement tasks and algorithm application files in the enterprise's local AI computing service platform. When processing the received algorithm arrangement tasks of the enterprise, the video receiving module 3 is notified according to the algorithm arrangement task information to pull the monitoring video from the enterprise video source 1. For videos that need to run the algorithm in real time 24 hours a day, the video stream is pulled 24 hours a day; for those that need to conduct video inspections in different time periods, the corresponding video stream is pulled during the required time period.

[0101] Step 3: At the same time, the enterprise-side AI analysis module 5 notifies the monitoring image receiving module 2 6 to receive the specified monitoring video image from the monitoring image sending module 12 on the group side according to the algorithm arrangement task received from the enterprise. The monitoring image sending module 12 on the group side receives the monitoring video images uploaded by other enterprises with insufficient computing power from the monitoring image storage module 9 on the group side, and then sends them to the monitoring image receiving module 2 6 on the enterprise side; the AI ​​analysis module 5 obtains the monitoring video images uploaded by other enterprises from the monitoring image receiving module 2 6 according to the task arrangement instruction, and loads the corresponding AI algorithm to analyze and process the received monitoring video images.

[0102] Step 4: The AI ​​analysis module 5 generates an alarm message for the problem found when executing the AI ​​analysis algorithm and sends it to the alarm receiving module 13 on the group side. The subsequent execution steps of the alarm receiving module 13, the alarm intelligent analysis module 10, and the computing power intelligent management and scheduling module 11 on the group side are the same as described above.

[0103] To further optimize the above technical solution, enterprises that report alarms for the first time need to be initialized. The initialization steps include:

[0104] Obtain computing power information of each enterprise, including computing resource information, surveillance video information, and required AI algorithm information;

[0105] Weight values ​​are assigned based on the computing power information of each enterprise, including assigning initial weight values ​​to the surveillance videos of each enterprise and assigning initial weight values ​​to the time period of each surveillance video; when assigning weight values, the highest AI computing resource utilization score is the goal.

[0106] Generate initial algorithm scheduling tasks based on weight values ​​and computing resource information.

[0107] In an exemplary embodiment, referring to Figure 4 , the specific steps are as follows:

[0108] Step 1: Each enterprise under the group reports its computing power resource information, monitoring video information, required AI algorithm information and other information to the group through the information reporting service of the computing power intelligent management and scheduling module 11 in the group-side AI algorithm and alarm management platform 7.

[0109] Step 2: The computing power intelligent management and scheduling module 11 assigns initial weight values ​​to the surveillance videos of each enterprise according to the information reported by the enterprise, and also assigns initial weight values ​​to the time periods of each surveillance video. For example, a larger weight value is assigned to the surveillance videos corresponding to important production workshops, and a larger weight value is assigned to the important production time periods of some workshop surveillance videos.

[0110] Step 3: The computing power intelligent management and scheduling module 11 gives an initial algorithm scheduling task for a certain enterprise based on the weight information and the computing power resource information of the enterprise. In the initial algorithm scheduling task, the surveillance videos that need to be monitored 24 hours a day are specified, and the surveillance videos that need to be polled in different time periods are specified.

[0111] Step 4: Based on the weight value of each surveillance video, the weight value of each time period in the surveillance video, the computing power resources of the enterprise, and certain algorithm rules, calculate the AI ​​computing resource utilization score of the algorithm orchestration task.

[0112] Step 5: Adjust the video numbers and quantity that need to run 24 hours a day in the task scheduling, adjust the time period for video polling that does not run 24 hours a day in the task scheduling, repeat steps 3 to 4, and finally select the algorithm scheduling task with the highest AI computing resource utilization score as the final initial algorithm scheduling task for the enterprise.

[0113] Step 6: Change to another enterprise and repeat steps 3 to 5 to select the initial algorithm orchestration tasks for all enterprises under the group.

[0114] Step 7: The group-side computing power intelligent management and scheduling module 11 sends the initial task scheduling instructions to each enterprise AI algorithm service platform, and sends the algorithms required by each enterprise to the enterprise. The enterprise-side AI analysis module saves the received initial algorithm scheduling tasks and algorithm application files on the enterprise's local AI computing service platform.

[0115] Implementation column 2:

[0116] A multi-level large-scale AI computing power intelligent scheduling method, comprising:

[0117] The enterprise side receives the surveillance video, generates alarm information through analysis, and uploads the alarm information to the group side;

[0118] The group updates the AI ​​algorithm to schedule tasks based on the alarm information and sends them to the enterprise.

[0119] The task of updating AI algorithm orchestration includes classifying enterprise computing power, including insufficient computing power and sufficient computing power; and collecting videos uploaded by enterprises with insufficient computing power and sending them to enterprises with sufficient computing power.

[0120] The specific implementation process is consistent with the aforementioned multi-level large-scale AI computing power intelligent scheduling system, so it will not be repeated here.

[0121] For large group-type enterprises with a large number of subordinate enterprises, the enterprise-side AI algorithm service platform performs 24-hour uninterrupted AI algorithm analysis and processing for the surveillance videos of important production sites in the enterprise that are prone to high risks in accordance with the algorithm arrangement task requirements issued by the group side; for the surveillance videos of other less important production sites in the enterprise, they are processed by video intelligent inspection, that is, only the surveillance videos in the time period prone to risks are subjected to AI algorithm analysis and processing. For enterprises with insufficient AI computing power, this method can use the enterprise AI computing power resources with maximum efficiency.

[0122] Furthermore, the enterprise-side AI algorithm service platform generates alarms for problems discovered during monitoring and reports them to the group. By comprehensively analyzing the alarm information data reported by all enterprises in the group, the probability of the monitored problems occurring in each monitoring channel and each time period of each enterprise is obtained. Combined with the weight information of the importance of each monitoring video of the enterprise, the weight information of the importance of the running AI algorithm, and the AI ​​computing power resource information of all enterprises under the group, the algorithm orchestration tasks with the highest total AI computing power resource utilization rate that include all enterprises are analyzed and selected from the whole group level. For enterprises with abundant AI computing power resources, the algorithm orchestration tasks of the enterprises that are finally selected will output computing power to enterprises with insufficient computing power resources. From the whole group level, this solution can use the AI ​​computing power resources of all enterprises with the highest efficiency.

[0123] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0124] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-level large-scale AI computing power intelligent scheduling system, characterized in that: Including the AI ​​algorithm and alarm management platform on the group side and the AI ​​algorithm service platform on the enterprise side; AI algorithm service platform, which is used to receive surveillance videos, generate alarm information through analysis, and upload the alarm information to the AI ​​algorithm and alarm management platform; The AI ​​algorithm and alarm management platform is used to update the AI ​​algorithm orchestration tasks based on the alarm information and send them to the AI ​​algorithm service platform.

2. The multi-level large-scale AI computing power intelligent scheduling system according to claim 1 is characterized in that: The AI ​​algorithm and alarm management platform includes an alarm receiving module, an alarm intelligent analysis module, and a computing power intelligent management and scheduling module; An alarm receiving module is used to receive alarm information and send it to the alarm intelligent analysis module; The alarm intelligent analysis module is used to analyze the alarm information to obtain the alarm frequency of the surveillance videos of all production sites of each enterprise, and the alarm frequency of each production site surveillance video in each period; The intelligent computing power management and scheduling module is used to update the AI ​​algorithm to schedule and distribute tasks based on the frequency and computing power resource information of each enterprise.

3. The multi-level large-scale AI computing power intelligent scheduling system according to claim 2 is characterized in that: The AI ​​algorithm service platform includes an AI analysis module and a video receiving module; Video receiving module, used to receive surveillance video data and send it to the AI ​​analysis module; The AI ​​analysis module is used to receive and analyze the monitoring video data according to the AI ​​algorithm orchestration tasks issued by the computing power intelligent management and scheduling module to generate alarm information, and send the alarm information to the alarm receiving module.

4. The multi-level large-scale AI computing power intelligent scheduling system according to claim 2 is characterized in that: The AI ​​algorithm and alarm management platform also includes a monitoring image receiving module 1 and a monitoring image sending module; Monitoring image receiving module 1, used to receive monitoring videos to be processed in enterprises with insufficient computing power; The monitoring image sending module is used to receive and send the surveillance videos to be processed to enterprises with sufficient computing power according to the AI ​​algorithm scheduling tasks.

5. The multi-level large-scale AI computing power intelligent scheduling system according to claim 4 is characterized in that: The AI ​​algorithm and alarm management platform also includes a monitoring image storage module, which is used to store the monitoring video to be processed received by the monitoring image receiving module 1, and send the monitoring video to be processed to the monitoring image sending module.

6. The multi-level large-scale AI computing power intelligent scheduling system according to claim 4 is characterized in that: The AI ​​algorithm service platform also includes a monitoring image interception and transmission module and a monitoring image receiving module 1; A monitoring image interception and transmission module is used to receive the monitoring video sent by the video receiving module and upload it to the monitoring image receiving module 1; The monitoring image receiving module 2 is used to receive the monitoring video to be processed sent by the monitoring image sending module and send it to the AI ​​analysis module.

7. The multi-level large-scale AI computing power intelligent scheduling system according to claim 2 is characterized in that: The computing power intelligent management and scheduling module is equipped with an information reporting service, which is used to collect the computing power information of the enterprise for initialization; the initialization steps include: Obtain computing power information of each enterprise, including computing resource information, surveillance video information, and required AI algorithm information; Allocate weight values ​​according to the computing power information of each enterprise, including allocating initial weight values ​​to the surveillance videos of each enterprise and allocating initial weight values ​​to the time period of each surveillance video; Generate initial algorithm scheduling tasks based on weight values ​​and computing resource information.

8. The multi-level large-scale AI computing power intelligent scheduling system according to claim 7 is characterized in that: When assigning weight values, the goal is to maximize the AI ​​computing resource utilization score.

9. A multi-level large-scale AI computing power intelligent scheduling method, characterized in that: The multi-level large-scale AI computing power intelligent scheduling system according to any one of claims 1 to 8 comprises: The enterprise side receives the surveillance video, generates alarm information through analysis, and uploads the alarm information to the group side; The group updates the AI ​​algorithm to schedule tasks based on the alarm information and sends them to the enterprise.

10. The multi-level large-scale AI computing power intelligent scheduling method according to claim 9 is characterized in that: The task of updating AI algorithm orchestration includes classifying enterprise computing power, including insufficient computing power and sufficient computing power; and collecting videos uploaded by enterprises with insufficient computing power and sending them to enterprises with sufficient computing power.

Images