Computing network cooperative scheduling method and device

Abstract

The application provides a kind of algorithm network cooperative scheduling method and device, it is related to data processing technical field, comprising: obtaining the computing power resource under rail transit environment, network resource and the service information containing service priority;Based on the computing power resource, the network resource and the service priority, the comprehensive scheduling score is calculated, and according to the comprehensive scheduling score, target scheduling mode is determined;According to the target scheduling mode, the computing power resource and the network resource are cooperatively allocated to process the service information.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to a computing network collaborative scheduling method and device. BACKGROUND

[0002] With the upgrading of rail transit intelligence, the business of vehicle-mounted video, danger identification and the like is booming, the daily data volume of a single train exceeds 2TB, and the business presents obvious tidal effect; at the same time, toB control type and toC consumption type businesses are mixed in the same 5G network, which puts forward differentiated requirements for low latency, high reliability and large bandwidth. The existing centralized processing mode causes a large amount of idle computing power on the vehicle, high end-to-end latency, and static QoS strategy cannot be dynamically adjusted according to space-time changes, so the resource utilization is low and the key business is easily disturbed, therefore, how to more effectively perform computing network collaborative scheduling has become a problem to be solved in the industry. SUMMARY

[0003] The present application provides a computing network collaborative scheduling method and device to solve the problem of how to more effectively perform computing network collaborative scheduling in the prior art.

[0004] The present application provides a computing network collaborative scheduling method, comprising:

[0005] Obtaining computing power resources, network resources and business information containing business priority in a rail transit environment;

[0006] Based on the computing power resources, the network resources and the business priority, a comprehensive scheduling score is calculated, and a target scheduling mode is determined according to the comprehensive scheduling score;

[0007] According to the target scheduling mode, the computing power resources and the network resources are collaboratively allocated to process the business information.

[0008] According to the computing network collaborative scheduling method provided by the present application, the target scheduling mode is determined according to the comprehensive scheduling score, comprising:

[0009] In the case that the comprehensive scheduling score is greater than a first preset threshold, the target scheduling mode is determined as a high-fidelity mode; wherein the high-fidelity mode is used to allocate dedicated network slice resources for the business corresponding to the business information;

[0010] In the case that the comprehensive scheduling score is less than a second preset threshold, the target scheduling mode is determined as a cache mode, wherein the cache mode is used to temporarily store the data corresponding to the business information in the vehicle terminal.

[0011] According to the computing network collaborative scheduling method provided by the present application, the target scheduling mode is determined according to the comprehensive scheduling score, comprising:

[0012] determining that the target scheduling mode is a balanced mode when the comprehensive scheduling score is between the first preset threshold and the second preset threshold;

[0013] The balanced mode is used for performing compression processing on data corresponding to the service information by a vehicle-mounted terminal and performing quality enhancement processing on the compressed data by an edge computing node.

[0014] According to the algorithm network collaborative scheduling method provided by the application, the method further comprises:

[0015] Based on the space-time feature data of the rail transit environment, the running mode in which the rail transit environment is located is identified, and the running mode includes a peak mode, an off-peak mode or a flat peak mode.

[0016] According to the target scheduling mode and the running mode, the computing power resources and the network resources are collaboratively allocated.

[0017] According to the algorithm network collaborative scheduling method provided by the application, according to the target scheduling mode and the running mode, the computing power resources and the network resources are collaboratively allocated, which comprises:

[0018] When the running mode is the peak mode and the target scheduling mode is the high-fidelity mode, the dedicated network slice resource is allocated for the service information corresponding to the service in the available network resource outside the hard isolation network channel reserved for the preset highest priority service.

[0019] According to the algorithm network collaborative scheduling method provided by the application, according to the target scheduling mode and the running mode, the computing power resources and the network resources are collaboratively allocated, which comprises:

[0020] In the case that the running mode is the off-peak mode, a large-bandwidth network channel is allocated for executing the target scheduling mode.

[0021] The application further provides an algorithm network collaborative scheduling device, comprising the following modules:

[0022] The acquisition module is used for acquiring computing power resources, network resources and service information containing service priority in a rail transit environment.

[0023] The determination module is used for calculating a comprehensive scheduling score based on the computing power resources, the network resources and the service priority, and determining a target scheduling mode according to the comprehensive scheduling score.

[0024] The allocation module is used for collaboratively allocating the computing power resources and the network resources to process the service information according to the target scheduling mode.

[0025] The application provides an algorithm-network cooperative scheduling device, and the device is also used for:

[0026] When the comprehensive scheduling score is greater than a first preset threshold, the target scheduling mode is determined as a high-fidelity mode; wherein the high-fidelity mode is used for allocating a dedicated network slice resource for a service corresponding to the service information;

[0027] When the comprehensive scheduling score is less than a second preset threshold, the target scheduling mode is determined as a cache mode, wherein the cache mode is used for temporarily storing data corresponding to the service information in a vehicle terminal.

[0028] The application provides an algorithm-network cooperative scheduling device, and the device is also used for:

[0029] When the comprehensive scheduling score is between the first preset threshold and the second preset threshold, the target scheduling mode is determined as a balanced mode;

[0030] The balanced mode is used for performing compression processing on the data corresponding to the service information by the vehicle terminal, and performing quality enhancement processing on the compressed data by an edge computing node.

[0031] The application provides an algorithm-network cooperative scheduling device, and the device is also used for:

[0032] Based on space-time feature data of a rail transit environment, an operation mode of the rail transit environment is identified, and the operation mode includes a peak mode, an off-peak mode or a flat-peak mode;

[0033] According to the target scheduling mode and the operation mode, the computing power resource and the network resource are cooperatively allocated.

[0034] The application provides an algorithm-network cooperative scheduling device, and the device is also used for:

[0035] When the operation mode is the peak mode and the target scheduling mode is the high-fidelity mode, a dedicated network slice resource is allocated for a service corresponding to the service information in available network resources outside a hard-isolated network channel reserved for a preset highest priority service.

[0036] The application provides an algorithm-network cooperative scheduling device, and the device is also used for:

[0037] When the operation mode is the off-peak mode, a large-bandwidth network channel is allocated for executing the target scheduling mode.

[0038] The application further provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the algorithm network cooperative scheduling method according to any one of the above when executing the computer program.

[0039] The application further provides a non-transitory computer readable storage medium, which stores a computer program, wherein the computer program is executed by a processor to implement the algorithm network cooperative scheduling method according to any one of the above.

[0040] The application further provides a computer program product, which includes a computer program, wherein the computer program is executed by a processor to implement the algorithm network cooperative scheduling method according to any one of the above.

[0041] The algorithm network cooperative scheduling method and device provided by the application continuously collect computing power resources, network resources and service information carrying service priority in the same control plane under the rail transit environment, establish a comprehensive scheduling scoring model with multi-dimensional real-time data as input, and select the target scheduling mode according to the dynamic change of the score, and then allocate and arrange the computing power and network resources. The process converts the original static and fragmented resource configuration into a closed-loop adaptive mechanism, so that the computing power between the vehicle terminal, the edge and the center cloud can flow as needed, and the network bandwidth can converge or release in real time according to the service urgency, thereby synchronously solving the problems of high idle rate of vehicle computing power, excessive end-to-end delay and mutual occupation of key services and non-key services, and providing intelligent scheduling capability with high reliability, low delay and significantly improved resource utilization for rail transit. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the application or prior art, the following will briefly introduce the drawings needed in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0043] Figure 1 is a flowchart of the algorithm network cooperative scheduling method provided by the application;

[0044] Figure 2 is a schematic diagram of the overall architecture provided by the application;

[0045] Figure 3 is a schematic diagram of the algorithm network cooperative scheduling device structure provided by the application;

[0046] Figure 4 is a schematic diagram of the structure of the electronic device provided by the application. DETAILED DESCRIPTION

[0047] In order to make the objects, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below with reference to the drawings in the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0048] Figure 1 is a flowchart of the algorithm network cooperative scheduling method provided by the present application, as shown in Figure 1 The method comprises the following steps:

[0049] In step 110, the computing power resource, network resource and service information containing service priority in the rail transit environment are obtained.

[0050] In the present application, the rail transit environment is a continuous mobile scenario with train high-speed operation along fixed tracks as the core, covering stations, tunnels, vehicle depots and control centers. In this scenario, heterogeneous services such as train control, video monitoring and passenger service run simultaneously, generating not only millisecond-level low-latency, ultra-high-reliability critical instruction streams, but also high-definition video streams. Moreover, the passenger flow fluctuates several times between peak and off-peak hours, resulting in a highly dynamic, load disparity and significant priority difference in the time and space dimensions of the multi-level computing power of the on-board terminal, trackside edge node and central cloud, as well as the 5G private network, public network and wired transmission network resources.

[0051] The computing power resource in the rail transit environment refers to the available capacity of the three-level computing facilities distributed in the on-board terminal, trackside edge node and central cloud in the train operation process: the on-board terminal periodically collects the instantaneous utilization rate of CPU, GPU, NPU, memory remaining capacity, battery capacity and remaining capacity of solid state disk through the operating system interface, and encapsulates these data as lightweight messages for reporting.

[0052] The trackside edge node also records the CPU / GPU occupancy rate, GPU temperature, allocatable container instance number and cache capacity of its own server node in a containerized manner; the central cloud aggregates the real-time load of virtual machines, bare metals and acceleration cards to form a global computing power pool view.

[0053] The network resource covers the real-time performance indicators of the 5G base station along the train line, the transmission network and the core network: the base station side continuously outputs the reference signal receiving power, the signal-to-noise ratio, the PDCP layer packet loss rate and the uplink and downlink available throughput according to the 5G NR measurement report; the transmission network reads the link delay, the FlexE hard slice residual bandwidth, the queue length and the congestion mark through the SRv6 controller; the core network side records the end-to-end delay, the jitter, the packet loss and the available bandwidth, and all indicators enter the algorithm network sensing plane in the form of streaming data.

[0054] The service information containing the service priority is the service demand description issued by the management platform or the vehicle-mounted application, and the main body is a three-level priority division: the emergency class service requires that the end-to-end delay does not exceed two hundred milliseconds and is assigned the highest priority, the critical class service allows a delay of two seconds or less and is assigned a medium priority, and the ordinary class service can be completed in five minutes or even longer and corresponds to the lowest priority; the information also carries the service type identifier, the required data fidelity level, the single data volume, the backhaul period and the maximum interruption time that can be tolerated, and after being encapsulated by RESTful API or a special protocol, it is parsed by the algorithm network service sensing engine and bound to the corresponding service flow, thereby forming the complete input required in step 110 together with the computing power and the network state.

[0055] In step 120, based on the computing power resource, the network resource and the service priority, a comprehensive scheduling score is calculated, and a target scheduling mode is determined according to the comprehensive scheduling score.

[0056] In the present application, the computing power resource, the network resource and the service information carrying the service priority are input into a dynamically weighted scoring model.

[0057] The dynamically weighted scoring model takes network quality, computing power load and service priority as factors, each of which is mapped to a sub-score Nscore, Cscore and Pscore in the range of 0-100, and is linearly superimposed according to the preset weights 0.6, 0.3 and 0.1 to generate a comprehensive scheduling score S in the range of 0-100.

[0058] The scoring result is refreshed in real time, and the target scheduling mode is determined according to the comprehensive scheduling score: when the comprehensive scheduling score S is greater than or equal to 80, the high-fidelity mode is triggered, and a dedicated network slice is allocated for the service; when the comprehensive scheduling score 60≤S<80, the balanced mode is enabled, and data compression and super-resolution reconstruction are realized; when the comprehensive scheduling score S is less than 60, the cache mode is switched to, and the data is temporarily stored in the vehicle-mounted SSD and is backhauled when the network is idle.

[0059] The dynamic weighted scoring model takes the network quality, the computing power margin and the service priority collected by the rail transit in real time as inputs, converts the RSRP, the SINR and the packet loss rate into network scores, maps the CPU, the GPU and the memory occupation rate into computing power scores, directly assigns scores to the services according to T1, T2 and T3, then linearly superimposes the fixed weights of 0.6, 0.3 and 0.1 to form a comprehensive score S between 0 and 100; the score is refreshed every second based on the sliding window average, and the threshold is adaptively adjusted when the network load is continuously lower or higher than a certain threshold for a long time, so that the system can always determine the subsequent real-time fidelity, compression enhancement or cache delay scheduling strategy according to the unique and quantifiable S value under the conditions of high-speed movement and tidal passenger flow, and realize the collaborative optimization of computing power and network resources.

[0060] In step 130, the computing power resources and the network resources are allocated in coordination according to the target scheduling mode to process the service information.

[0061] In the present application, if the mode is a high-fidelity mode, the scheduling center first calls the network management engine, reserves a FlexE hard slice for the service along the SRv6 path, and locks the bandwidth and time slot at the base stations, transmission nodes and core network along the way at the same time, so as to ensure that the end-to-end delay is locked within 10ms; at the same time, the inference or transcoding task is directly mapped to the edge GPU instance closest to the user by the computing power offloading engine, the container image is second-level issued through KubeEdge, and only the result is returned after the task is completed at the edge, so as to eliminate the round-trip delay of the cloud.

[0062] If the mode is an equalization mode, the system first starts a lightweight encoding container on the vehicle-mounted terminal NPU, compresses the original video at a target code rate of 50%, and sends the compressed code stream to the trackside MEC through the existing 5G bearer; the MEC runs the super-resolution reconstruction model on the GPU in the same cabinet to restore the 720p stream to 1080p visual quality, and then sends it to the central platform through the ordinary backhaul link, so that the network occupancy is reduced by half while the picture quality is lossless.

[0063] If the mode is a cache mode, the computing power offloading engine immediately writes the to-be-transmitted data into the vehicle-mounted SSD array, and sends a "delayed scheduling" instruction to the network management engine to release the air interface bandwidth; when the train enters the terminal station and the network utilization rate is continuously lower than 40% for five minutes, the scheduling center triggers the batch backhaul task again, uses the 200% peak bandwidth channel during the idle time to upload the cached data completely, and ensures the data integrity and the optimal filling of network resources.

[0064] In the present application, by continuously collecting computing power resources, network resources and service information carrying service priority in the same control plane in the rail transit environment, a comprehensive scheduling score model is established with multi-dimensional real-time data as input; according to the dynamic changes of the score, the system selects the adaptive target scheduling mode in real time without manual intervention, and accordingly allocates and arranges the computing power and network resources. This process converts the originally static and fragmented resource configuration into a closed-loop adaptive mechanism, so that the computing power between the vehicle terminal, the trackside edge and the center cloud can flow on demand, and the network bandwidth also converges or releases in real time according to the service urgency, thereby synchronously solving the existing problems of high idle rate of vehicle computing power, excessive end-to-end delay and mutual occupation of key services and non-key services, and providing intelligent scheduling capability with high reliability, low delay and significantly improved resource utilization for rail transit.

[0065] Optionally, according to the comprehensive scheduling score, a target scheduling mode is determined, comprising:

[0066] In the case where the comprehensive scheduling score is greater than a first preset threshold, the target scheduling mode is determined as a high-fidelity mode; wherein the high-fidelity mode is used to allocate dedicated network slice resources for the service information corresponding service.

[0067] In the case where the comprehensive scheduling score is less than a second preset threshold, the target scheduling mode is determined as a cache mode, wherein the cache mode is used to temporarily store the data corresponding to the service information in the vehicle terminal.

[0068] In the present application, in the rail transit running scene, when the comprehensive scheduling score is higher than the first preset threshold, the system determines that the wireless coverage of the section where the current train is located, the trackside edge node and the vehicle terminal are all in the state of abundant resources, and then the high-fidelity mode is started.

[0069] In the high-fidelity mode, the network side cooperates with the bearer network through the train dedicated 5G base station to establish an end-to-end hard isolation network slice for train control or driver behavior monitoring type services. This slice uses FlexE time slot exclusively on the base station side, and realizes hop-by-hop locking through SRv6 Segment Routing on the transmission network side, so as to ensure that the train communication delay is always less than 10 milliseconds and the jitter is not more than 2 milliseconds during the movement of the whole train at a speed of 80 to 120 kilometers per hour; at the same time, the vehicle computing unit unloads the corresponding video analysis task to the trackside MEC server of the nearest station, uses its GPU resources to complete real-time inference, and only returns the recognition result to the control center, thereby avoiding the delay caused by long-distance cloud side round trip, and meeting the time limit requirements of train emergency braking or driver abnormal behavior alarm.

[0070] If the comprehensive scheduling score is lower than the second preset threshold, the system identifies that the current train is entering a tunnel or a weakly overlapped coverage area of a base station, and that the network instantaneous capacity is insufficient, and thus enters the buffering mode.

[0071] In the buffering mode, the vehicle-mounted gateway immediately writes high-definition video data collected by the vehicle-mounted camera in the form of file blocks into an NVMe solid-state storage array in a train cabinet in sequence, and writes a train number, a carriage number and a UTC timestamp in a file index, so as to ensure that the data correspond to a vehicle running track; meanwhile, the network side releases air interface resources occupied by the service, and preferentially preserves a CBTC signal system and an emergency call service.

[0072] When the train runs to a vehicle depot or a terminal station, and the network utilization rate is lower than 40% for five consecutive minutes, the system uses a special large-bandwidth channel of the vehicle depot night maintenance window to batch return the buffered data to the center cloud, completes data integrity verification, and the solid-state storage array automatically recovers space, ready for the next round of operation cycle.

[0073] In the present application, in the train operation, high-priority services always enjoy a dedicated network slice, low-priority services are automatically buffered and batch returned in idle time, the utilization rate of vehicle-mounted computing power is improved, the network peak bandwidth demand is reduced, and the reliability of safety services is ensured.

[0074] Optionally, the target scheduling mode is determined according to the comprehensive scheduling score, including:

[0075] When the comprehensive scheduling score is between the first preset threshold and the second preset threshold, the target scheduling mode is determined as the balanced mode.

[0076] The balanced mode is used for performing compression processing on data corresponding to the service information by the vehicle-mounted terminal, and performing quality enhancement processing on the compressed data by the edge computing node.

[0077] When the comprehensive scheduling score falls between the first and second preset thresholds, the system determines that the current network and computing power are in a usable but tense state, and then enables the balanced mode.

[0078] In the balanced mode, the vehicle-mounted terminal immediately calls the NPU to perform lightweight encoding on the original video, compresses the code rate by about 50%, and encapsulates into a low-latency frame stream; the low-latency frame stream is sent to the trackside MEC of the nearest station through the 5G train-ground link, and the GPU in the MEC completes super-resolution reconstruction within 30ms, restores the 720p picture to 1080p visual quality, and then sends it to the center platform through the existing backhaul channel. While the process reaches the key frame within 30ms, the air interface bandwidth occupancy is halved, which not only relieves wireless congestion during peak hours, but also fully utilizes the idle computing power of vehicles and edges, achieving balanced optimization of latency, bandwidth and picture quality.

[0079] Optionally, the method further comprises:

[0080] Based on the space-time feature data of the rail transit environment, an operation mode in which the rail transit environment is located is identified, and the operation mode comprises a peak mode, an off-peak mode or a flat-peak mode.

[0081] According to the target scheduling mode and the operation mode, the computing power resources and the network resources are allocated in coordination.

[0082] In the present application, during the train operation, the system continuously collects the space-time feature data of the rail transit environment, i.e. the current train line section, time, station passenger flow density and historical passenger flow curve, and accordingly divides the overall operation state into a peak mode, a flat-peak mode or an off-peak mode.

[0083] The dispatching hub performs two-dimensional matching of the determined target scheduling mode and the operation mode: when the train is located in a transfer hub and is in the morning or evening peak period, the system first reserves a FlexE hard isolation channel for the highest priority business such as CBTC signal and driver monitoring, and then performs the current target mode in the remaining bandwidth; if the train enters the vehicle depot and is in the night off-peak time, a dedicated backhaul channel with 200% peak bandwidth is opened, and a software upgrade package or batch of backhaul inspection data is issued in advance;

[0084] In the flat-peak period, the key frame is put into the TSN deterministic channel and the non-key frame shares the elastic bandwidth to maximize the QoS / energy consumption ratio, so as to realize the coupling optimization of the operation mode and the scheduling strategy, so that the computing power, storage and network resources are accurately put in the time and space dual dimensions, and the coexistence of peak resource gap and flat-peak redundancy is avoided.

[0085] Optionally, according to the target scheduling mode and the operation mode, the computing power resources and the network resources are allocated in coordination, which comprises:

[0086] When the operation mode is the peak mode and the target scheduling mode is the high-fidelity mode, a dedicated network slice resource is allocated for the business corresponding to the business information in the available network resources outside the hard isolation network channel reserved for the preset highest priority business.

[0087] In the present application, in the peak mode, the train runs in the transfer hub and is accompanied by large passenger flow, and the system first locks a FlexE hard isolation channel for the highest priority business such as CBTC signal and driver monitoring, and the channel is configured in a fixed time slot exclusively along the base station, transmission node and core network, so as to ensure that the time delay and jitter are always within the safety threshold.

[0088] Subsequently, the dispatching center establishes a dedicated network slice for the next high priority service corresponding to the current service information in the remaining network resources outside the hard isolation channel under the triggering condition of the high-fidelity mode.

[0089] The slice completes path arrangement in milliseconds through SRv6 segment routing, and the bandwidth size is dynamically adjusted according to real-time scoring, which neither competes for resources with train control instructions nor provides predictable low-latency transmission for the service during peak congestion periods, thereby realizing differentiated and reliable fidelity of multi-level services in peak scenarios.

[0090] Optionally, the computing resource and the network resource are allocated in coordination according to the target scheduling mode and the running mode, including:

[0091] In the case where the running mode is an idle mode, a large bandwidth network channel is allocated for executing the target scheduling mode.

[0092] In the idle mode of train return section or night full-line shutdown, the dispatching center converges the idle FlexE time slots and optical layer channels on the whole transmission link into a large bandwidth network channel at one time, and the capacity of the channel can reach 200% of the daytime peak.

[0093] The channel uses 100Gbps port bundling at the physical layer, and is marked as an “idle dedicated” slice in the SRv6 controller, and is only open to low real-time services such as cache backhaul, software upgrade, and device inspection data.

[0094] The on-board terminal, trackside MEC, and central cloud upload the previously cached high-definition video, log file, or AI model incremental package in a multi-path parallel manner through the channel, and the average throughput is improved to 1.2Gbps, which can complete 200GB level data backhaul in the maintenance window of 02:00-04:00, and then the channel is automatically released, ensuring that the resources are zero before the next day peak, and realizing full utilization of idle network resources and precise scheduling of computing network coordination.

[0095] In an optional embodiment, Figure 2 The overall framework provided by the present application is shown in the following figure: Figure 2As shown, the scheme analyzes the rail transit user business type and real-time requirements. On the one hand, the computing power and network synergy advantages are fully utilized, part of the computing power tasks are offloaded to the end side, the end computing gateway computing power is combined to perform video on-demand compression coding during peak period, the pressure on network bandwidth is reduced, and user experience is improved. On the other hand, a scheduling system is constructed, according to traffic prediction, business index requirements, etc., dynamic adaptive transmission and processing strategies are used at different time periods and different stations, which can meet the requirements of related industries on video fidelity on the basis of ensuring real-time processing, and realize the balance of delay and quality. Specifically, it includes an algorithm network resource access component, an algorithm network dynamic perception component, and an algorithm network intelligent scheduling component.

[0096] The algorithm network resource access component mainly collects raw data of algorithm network resources through basic resource access (cloud, edge, end, network, etc.), adopts a hybrid access mode, configures Kafka streaming collection channels for time series data such as signal systems and video monitoring, uses Fluentd log collection framework for unstructured data such as device logs, supports management of collection data period and mode, and pre-processing, adaptation, conversion, etc. for multiple access data. At the same time, through the interface of the end computing gateway and the cloud platform, video processing, AI inference, data processing and other tasks are flexibly assigned to related computing resources, effectively improving the utilization rate of vehicle terminal computing power resources, and integrating SRv6 programmable forwarding technology and FlexE hard slicing technology to provide deterministic delay fidelity for rail transit key services, improve network resource utilization, and reduce conflict rate of different priority services. Specifically, it includes a basic resource interface engine, a computing power offloading engine, and a network management engine.

[0097] Further, the basic resource interface engine interfaces with various resources such as system bottom cloud, edge, end, and network, and manages the bottom basic resources through interface acquisition, active reporting, and intermediate table acquisition. The interface acquisition supports passive acquisition of computing power information through the interface; the active reporting supports active opening of reporting interface and active reporting of computing power information through the interface; the intermediate table acquisition supports providing data table format and periodically reporting data in fixed format.

[0098] Further, the computing power offloading engine flexibly assigns video processing, AI inference, data processing and other tasks to related computing resources through the interface of the end computing gateway and the cloud platform, effectively improves the utilization rate of vehicle terminal computing power resources, and at the same time reduces the load of the cloud and the transmission pressure of the network.

[0099] The end computing gateway is deployed in a train on-board terminal and a station edge device, supports a lightweight container runtime environment, realizes virtualization and encapsulation of local computing power resources, the gateway is built-in with a priority queue management algorithm, monitors CPU / GPU / NPU computing power margins in real time, and reports resource states through an MQTT protocol.

[0100] The cloud platform interface builds a bidirectional communication channel, supports access of heterogeneous resource pools across cloud service providers, adopts adaptive protocol conversion technology in the interface layer, and realizes seamless connection of Kubernetes cluster APIs and track transportation special protocols.

[0101] For example, for time-sensitive AI inference tasks, a model segmentation technology is used to deploy a feature extraction layer in a terminal NPU, thereby reducing network transmission pressure.

[0102] Further, the network management engine integrates SRv6 programmable forwarding technology through a network scheduling interface and a network resource allocation management module, realizes dynamic scheduling of cross-base station service chains, and integrates FlexE hard slicing technology to provide deterministic delay fidelity for track transportation key services, thereby realizing intelligent allocation of network resources under cooperation of toB (enterprise services, including train control and device monitoring) and toC (consumer services, including passenger services and commercial applications) services, improving network resource utilization, and reducing conflict rates of different priority services.

[0103] The algorithm network dynamic perception component perceives underlying computing power resources, network resources and storage resources on one hand, builds a terminal-edge two-level computing power evaluation system, and monitors available computing resources (CPU / GPU / NPU utilization, memory capacity, battery status) of mobile terminals (on-board devices / handheld terminals) in real time; on the other hand, perceives user-side latency, transmission data volume, upload traffic, task priority, data fidelity requirements (4K / 1080P / 720P), and evaluates current service quality. Specifically, the algorithm network dynamic perception component includes an algorithm network resource perception engine and an algorithm network service perception engine.

[0104] Further, the algorithm network resource perception engine mainly integrates perception capabilities of computing power (terminal computing power, edge computing power, industry cloud), network and service application, uniformly collects and manages performance parameters of network devices such as 5G base stations, various perception data sources of cloud management platforms, and realizes perception of network performance and resource load in view of typical algorithm network business scenarios of the track transportation industry. The algorithm network resource perception engine includes computing power resource perception and network resource perception.

[0105] The computing power resource perception mainly uses Linux cgroup technology to collect computing resource CPU occupancy, GPU temperature, storage resource SSD remaining capacity, and IOPS in real time.

[0106] The network resource perception is mainly based on 5G NR measurement report to obtain RSRP, SINR and PDCP layer packet loss rate.

[0107] Further, the algorithm network service perception engine perceives user side latency, transmission data volume, upload traffic, and task priority, data fidelity requirement and other indicators, analyzes and processes data according to the indicator system, constructs a user service classification model, supports real-time decision-making of algorithm network scheduling and operation fidelity, intelligent operation and maintenance and other demand scenarios. The user service classification model includes:

[0108] Emergency class (T1 level): driver dangerous behavior recognition (latency requirement ≤200ms);

[0109] Key class (T2 level): equipment fault diagnosis (allowable latency ≤2s);

[0110] Normal class (T3 level): carriage monitoring video return (allowable latency ≥5 minutes);

[0111] The algorithm network intelligent scheduling component mainly forms the track traffic service priority scheduling, time-sharing tidal scheduling and other capabilities through system modeling, business rule analysis, optimization strategy solving and system module docking, constructs the algorithm network collaborative intelligent scheduling system for track traffic application, including the hierarchical scheduling strategy of real-time fidelity mode, end algorithm power compression transmission + track traffic local enhancement mode, suspension transmission + local cache mode for different priority task scenarios, and the dynamic adaptive transmission and processing strategy for different time periods and different stations for tidal effect business scenarios, which effectively improves the efficiency of algorithm resource utilization, reduces the network transmission pressure, improves the user experience, and specifically includes the track traffic service priority scheduling engine and the track traffic time-sharing tidal scheduling engine.

[0112] Further, the track traffic service priority scheduling engine faces the scenarios of track traffic video and other tasks with different priority levels of emergency class, key class and normal class, intelligently analyzes and decides through system modeling, algorithm optimization and strategy generation according to the perceived algorithm network resource indicators and business operation indicators, and selects appropriate algorithm nodes, routing strategies and bandwidth allocation strategies to provide services. It includes the following steps:

[0113] The user issues a video collection task from the management platform and determines the business quality indicators such as latency, establishes a dynamic weighted scoring model combined with algorithm, network and other resource indicators, calculates the strategy score based on network quality (RSRP / SINR), terminal algorithm load and business priority, and provides a basis for selecting the optimal transmission strategy.

[0114] ;

[0115] Wherein, .

[0116] Network score : calculated according to RSRP / SINR;

[0117] Computing power score : calculated based on terminal CPU / GPU / NPU utilization and edge node load;

[0118] Service score : T1=100, T2=60, T3=20.

[0119] The scheduling platform performs hierarchical scheduling strategy execution according to the dynamic weighted score, and fully schedules the terminal, edge computing power and network, respectively using real-time fidelity mode, terminal computing power compression transmission + rail transit local enhancement mode, suspension transmission + local cache mode and other strategy modes. The score interval is shown in Table 1 as follows:

[0120] Table 1

[0121]

[0122] High-fidelity mode, mainly for T1 high-priority services (such as driver behavior monitoring, etc.) to allocate dedicated network slices;

[0123] Balanced mode, i.e. terminal computing power compression transmission + rail transit local enhancement mode, mainly for T2 medium-priority service video stream to implement hierarchical coding (code rate compression 50%) and off-track MEC node to perform super-resolution reconstruction;

[0124] Cache mode, i.e. suspension transmission + local cache mode, mainly stores T3 low-priority data to the on-board SSD array when the network is congested. For example, in the case of network congestion, real-time monitoring video data is temporarily transmitted, or in the form of terminal computing power offloading, part of the preprocessed low-fidelity data is transmitted, so that the fidelity of driver dangerous behavior monitoring and other high-priority data is ensured, so as to realize the collaborative scheduling of computing power and network.

[0125] Based on the global scheduling management strategy, when the network is idle (network utilization <40% and lasts for ≥5 minutes) or the train enters the terminal station maintenance area (GPS coordinates match), the local fidelity data is temporarily stored for uploading to ensure data integrity.

[0126] According to the changes of the perceived network quality and resource load and other indicators, the current service scheduling strategy is dynamically adjusted to form a feedback optimization closed loop.

[0127] Further, the rail transit time-sharing tidal scheduling engine is mainly aimed at the business scenarios of rail transit during rush hours and the like with tidal effects, coordinates the supply and scheduling of distributed computing power, storage, and traffic, improves the application of end computing power, adopts dynamic adaptive transmission and processing strategies at different time periods and different stations, and performs end-to-end monitoring on the scheduling execution process, automatically processes exceptions and errors according to relevant strategies, can meet the requirements of related industries on video fidelity and the like on the basis of ensuring the real-time processing, reduces the network peak bandwidth pressure, and improves the customer experience. The method comprises the following steps:

[0128] The rail transit space-time characteristics of different time periods and different stations are analyzed, and divided into three modes of peak, flat peak, and idle time, as shown in the following table 2.

[0129] Table 2

[0130]

[0131] Different modes are matched with differentiated and dynamic adaptive resource allocation and optimized scheduling strategies. Specifically as follows:

[0132] a) In the peak mode (S≥80), a hard isolated channel (FlexE slice≥800Mbps) is reserved for the train control system, which is only used for critical traffic such as train emergency braking and ATO instructions, and ensures the direct transmission of SRv6 key business (latency≤10ms); and dynamic bandwidth preemption is adopted, when the control instruction delay exceeds 15ms, the passenger commercial traffic bandwidth is automatically reduced by a certain proportion (maximum release of 70%), and the millisecond-level resource reconfiguration is realized through SRv6 segment routing, to ensure the absolute reliability and real-time performance of high-priority services such as train control and emergency communication.

[0133] b) In the flat peak mode (40≤S<80), the maximum comprehensive performance ratio (QoS / energy consumption) is taken as the target, the key frame is allocated with a TSN deterministic channel (latency≤30ms), and the non-key frame uses elastic shared bandwidth, to maximize the resource utilization efficiency on the premise of ensuring the basic service quality.

[0134] c) In the idle time mode (S<40), a dedicated channel (bandwidth up to 200% of the daily peak) is opened at 02:00-04:00 every day for batch transmission of equipment inspection data, and intelligent pre-distribution of software upgrade is carried out using idle network resources, 80% of the upgrade package is issued in advance, and the remaining 20% of the incremental update is completed before the peak.

[0135] A strategy knowledge base is established to store verified and effective scheduling schemes, and the scheduling strategy is continuously improved according to the changes of resource indicators and business indicators, to meet the needs of different business scenarios of rail transit.

[0136] The scheme of the application can, according to network resource indexes, service index requirements and the like, adopt hierarchical scheduling strategies such as a real-time fidelity mode, end-computing power compression transmission + local enhancement mode of rail transit, and suspension transmission + local caching mode for different priority rail transit task scenarios such as emergency, key and ordinary classes, and can adopt dynamic adaptive transmission and processing strategies for different time periods and different stations for rail transit tidal effect service scenarios, so as to meet the requirements of related services on video fidelity and the like on the basis of ensuring real-time processing, achieve a balance between delay and quality, and provide effective support for intelligent evolution of rail transit services.

[0137] Specifically, through the connection of end computing gateways and cloud platform interfaces, the video processing, AI inference and data processing tasks can be flexibly issued to related computing resources, the utilization rate of on-board terminal computing power resources is effectively improved by more than 30%, and the cloud load and network transmission pressure are reduced; through dynamic adaptive transmission and processing strategies for different time periods and different stations, the absolute reliability and real-time performance of high-priority services such as train control and emergency communication are ensured, and the service efficiency is improved by more than 50%.

[0138] The algorithm network cooperative scheduling device provided by the application will be described below. The algorithm network cooperative scheduling device described below can be correspondingly referred to the algorithm network cooperative scheduling method described above.

[0139] Figure 3 The structure diagram of the algorithm network cooperative scheduling device provided by the application is shown in Figure 3 , which includes:

[0140] The acquisition module 310 is configured to acquire computing power resources, network resources and service information containing service priorities in a rail transit environment.

[0141] The determination module 320 is configured to calculate a comprehensive scheduling score based on the computing power resources, the network resources and the service priorities, and determine a target scheduling mode according to the comprehensive scheduling score.

[0142] The distribution module 330 is configured to cooperatively distribute the computing power resources and the network resources to process the service information according to the target scheduling mode.

[0143] According to the algorithm network cooperative scheduling device provided by the application, the device is further configured to:

[0144] In the case where the comprehensive scheduling score is greater than a first preset threshold, the target scheduling mode is determined to be a high-fidelity mode; wherein the high-fidelity mode is used to allocate dedicated network slice resources for services corresponding to the service information.

[0145] When the comprehensive scheduling score is less than a second preset threshold, determining that the target scheduling mode is a cache mode, wherein the cache mode is used for temporarily storing data corresponding to the service information in the vehicle terminal.

[0146] According to the application, a kind of algorithm network collaborative scheduling device is provided, the device is also used for:

[0147] When the comprehensive scheduling score is between the first preset threshold and the second preset threshold, determining that the target scheduling mode is a balanced mode.

[0148] The balanced mode is used for compressing data corresponding to the service information by vehicle terminal, and quality enhancement processing is carried out on the compressed data by edge computing node.

[0149] According to the application, a kind of algorithm network collaborative scheduling device is provided, the device is also used for:

[0150] Based on the space-time feature data of rail transit environment, the running mode of the rail transit environment is identified, and the running mode includes peak mode, slack time mode or flat peak mode.

[0151] According to the target scheduling mode and the running mode, the computing power resource and the network resource are collaboratively allocated.

[0152] According to the application, a kind of algorithm network collaborative scheduling device is provided, the device is also used for:

[0153] When the running mode is peak mode, and the target scheduling mode is high-fidelity mode, in the available network resource outside the hard isolation network channel reserved for the preset highest priority service, dedicated network slice resource is allocated for the service information corresponding service.

[0154] According to the application, a kind of algorithm network collaborative scheduling device is provided, the device is also used for:

[0155] In the case where the running mode is slack time mode, large bandwidth network channel is allocated for executing the target scheduling mode.

[0156] This invention continuously collects computing resources, network resources, and service information carrying service priorities within the same control plane under the rail transit environment. Using this multi-dimensional real-time data as input, a comprehensive scheduling scoring model is established. Based on the dynamic changes in this score, the system instantly selects an appropriate target scheduling mode without manual intervention and accordingly coordinates and orchestrates the allocation of computing and network resources. This process transforms the originally static and fragmented resource allocation into a closed-loop adaptive mechanism, enabling the on-demand flow of computing power between onboard terminals, trackside edge computing, and the central cloud. Network bandwidth also converges or releases in real time according to the urgency of services. Thus, without changing the total amount of physical resources, it simultaneously solves existing problems such as high onboard computing power idle rate, excessive end-to-end latency, and mutual crowding between critical and non-critical services, providing rail transit with intelligent scheduling capabilities that are highly reliable, have low latency, and significantly improve resource utilization.

[0157] Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 4 As shown, the electronic device may include: a processor 410, a communication interface 420, a memory 430, and a communication bus 440, wherein the processor 410, the communication interface 420, and the memory 430 communicate with each other through the communication bus 440. The processor 410 can call logical instructions in the memory 430 to execute a computing network collaborative scheduling method, which includes: acquiring computing resources, network resources, and service information including service priorities in the rail transit environment;

[0158] Based on the computing power resources, the network resources, and the service priority, a comprehensive scheduling score is calculated, and a target scheduling mode is determined according to the comprehensive scheduling score.

[0159] According to the target scheduling mode, the computing resources and network resources are allocated in a coordinated manner to process the business information.

[0160] In addition, the logic instructions in the memory 430 described above can be implemented in the form of a software function unit and sold or used as an independent product, which can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0161] In another aspect, the present application also provides a computer program product, which comprises a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program is executed by a processor, so that the computer can execute the algorithm network cooperative scheduling method provided by the above-mentioned methods, and the method comprises the following steps: obtaining computing power resources, network resources and service information containing service priority in the rail transit environment;

[0162] Based on the computing power resources, the network resources and the service priority, a comprehensive scheduling score is calculated, and a target scheduling mode is determined according to the comprehensive scheduling score;

[0163] According to the target scheduling mode, the computing power resources and the network resources are cooperatively allocated to process the service information.

[0164] In another aspect, the present application also provides a computer program product, which comprises a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program is executed by a processor, so that the computer can execute the algorithm network cooperative scheduling method provided by the above-mentioned methods, and the method comprises the following steps: obtaining computing power resources, network resources and service information containing service priority in the rail transit environment;

[0165] Based on the computing power resources, the network resources and the service priority, a comprehensive scheduling score is calculated, and a target scheduling mode is determined according to the comprehensive scheduling score;

[0166] According to the target scheduling mode, the computing power resources and the network resources are cooperatively allocated to process the service information.

[0167] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purposes of the embodiments according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0168] Through the description of the above embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and the necessary general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0169] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A computer network collaborative scheduling method, characterized in that, include: Acquire computing resources, network resources, and business information including business priorities in the rail transit environment; Based on the computing power resources, the network resources, and the service priority, a comprehensive scheduling score is calculated, and a target scheduling mode is determined according to the comprehensive scheduling score. According to the target scheduling mode, the computing resources and network resources are coordinated to process the service information; The determination of the target scheduling mode based on the comprehensive scheduling score includes: If the overall scheduling score is greater than a first preset threshold, the target scheduling mode is determined to be a high-fidelity mode; wherein, the high-fidelity mode is used to allocate dedicated network slice resources for the service corresponding to the service information; When the comprehensive scheduling score is less than the second preset threshold, the target scheduling mode is determined to be a cache mode, wherein the cache mode is used to temporarily store the data corresponding to the service information in the vehicle terminal. The determination of the target scheduling mode based on the comprehensive scheduling score includes: When the comprehensive scheduling score is between the first preset threshold and the second preset threshold, the target scheduling mode is determined to be the balanced mode; The balanced mode is used to compress the data corresponding to the service information through the vehicle terminal, and to enhance the quality of the compressed data at the edge computing node.

2. The computer network collaborative scheduling method according to claim 1, characterized in that, The method further includes: Based on the spatiotemporal characteristic data of the rail transit environment, the operating mode of the rail transit environment is identified, including peak mode, off-peak mode or off-peak mode. Based on the target scheduling mode and the operating mode, the computing resources and the network resources are allocated in a coordinated manner.

3. The computer network collaborative scheduling method according to claim 2, characterized in that, Based on the target scheduling mode and the operating mode, the computing resources and network resources are allocated in a coordinated manner, including: When the operating mode is peak mode and the target scheduling mode is high-fidelity mode, dedicated network slice resources are allocated to the service corresponding to the service information from the available network resources outside the hard-isolated network channel reserved for the preset highest priority service.

4. The computer network collaborative scheduling method according to claim 2, characterized in that, Based on the target scheduling mode and the operating mode, the computing resources and network resources are allocated in a coordinated manner, including: When the operating mode is idle mode, a high-bandwidth network channel is allocated for executing the target scheduling mode.

5. A computer network collaborative scheduling device, characterized in that, include: The acquisition module is used to acquire computing resources, network resources, and business information including business priorities in the rail transit environment. The determination module is used to calculate a comprehensive scheduling score based on the computing power resources, the network resources, and the service priority, and to determine the target scheduling mode based on the comprehensive scheduling score; The allocation module is used to coordinately allocate the computing resources and the network resources according to the target scheduling mode to process the service information; The device is also used for: If the overall scheduling score is greater than a first preset threshold, the target scheduling mode is determined to be a high-fidelity mode; wherein, the high-fidelity mode is used to allocate dedicated network slice resources for the service corresponding to the service information; When the comprehensive scheduling score is less than the second preset threshold, the target scheduling mode is determined to be a cache mode, wherein the cache mode is used to temporarily store the data corresponding to the service information in the vehicle terminal. When the comprehensive scheduling score is between the first preset threshold and the second preset threshold, the target scheduling mode is determined to be the balanced mode; The balanced mode is used to compress the data corresponding to the service information through the vehicle terminal, and to enhance the quality of the compressed data at the edge computing node.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the computer network collaborative scheduling method as described in any one of claims 1 to 4.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the computer network collaborative scheduling method as described in any one of claims 1 to 4.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the computer network collaborative scheduling method as described in any one of claims 1 to 4.

Images