Intelligent virtualized resource dynamic scheduling system and method integrating urban governance data

Abstract

The present invention discloses an intelligent virtualized resource dynamic scheduling system and method that integrates urban governance data. The method includes: obtaining urban governance data and resource indicator data; predicting the GPU demand curve based on the urban governance data input into the event influence quantification model; wherein the event influence quantification model is obtained by training a dual-channel architecture model using historical GPU utilization data and corresponding urban governance data as a sample set, and the event influence quantification model includes a temporal convolutional network, a graph attention network, and a gated recurrent unit; generating a dynamic scheduling strategy; and executing resource scheduling. By implementing the method of the present invention, it is possible to accurately respond to specific events and high-load challenges in urban governance, ensure the service quality of key businesses, and solve the problems that the existing resource scheduling strategy ignores the comprehensive analysis of specific events and external data, resulting in a slow response when responding to instantaneous resource demand changes, affecting service quality and system stability.

Description

Technical Field

[0001] The present invention relates to a resource scheduling method, and more specifically to an intelligent virtualized resource dynamic scheduling system and method that integrates urban governance data. Background Art

[0002] In the context of modern urban governance, systems such as traffic management, emergency response, and environmental monitoring often face significant load fluctuations. These fluctuations are often triggered by specific events. For example, during heavy rainstorms, urban flooding monitoring systems urgently require the use of large amounts of GPU (Graphics Processing Unit) computing power for real-time image recognition to monitor and predict flood risks. However, under normal circumstances, the resource requirements of such systems are very low. Therefore, how to efficiently and flexibly schedule computing resources has become a key technical challenge.

[0003] Traditional resource scheduling strategies typically rely on historical load thresholds or real-time resource utilization to trigger adjustments. This approach is slow to respond to instantaneous changes in resource demand caused by specific events, easily causing service response delays or even system crashes. Current virtualization platforms typically focus only on their own resource monitoring data and fail to effectively integrate comprehensive analysis with urban governance event data such as weather warnings, traffic flow changes, and hotspots. This lack of cross-domain collaborative analysis capabilities limits the accuracy and timeliness of resource scheduling decisions. Existing time series prediction models are often based on historical resource usage data, ignoring the direct impact of external events on resource demand. For example, in the event of sudden rainstorms, resource demand forecasts will be significantly biased due to the lack of consideration of meteorological factors, affecting service quality. Although Chinese patent CN112187894A mentions a method for dynamic container scheduling based on load correlation prediction, it does not fully address the sudden changes in resource demand driven by specific events in urban governance scenarios and struggles to meet the high reliability requirements of smart city services.

[0004] Therefore, it is necessary to design a new method to accurately respond to specific events and high-load challenges in urban governance and ensure the service quality of key businesses, so as to solve the problems that the existing resource scheduling strategy ignores the comprehensive analysis of specific events and external data, resulting in slow response to instantaneous changes in resource demand, affecting service quality and system stability. Summary of the Invention

[0005] The purpose of the present invention is to overcome the defects of the prior art and provide an intelligent virtualized resource dynamic scheduling system and method that integrates urban governance data.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for dynamic scheduling of intelligent virtualized resources integrating urban governance data, comprising:

[0007] Obtain urban governance data and resource indicator data;

[0008] The GPU demand curve for a period of time in the future is predicted based on the input of the urban governance data into the event influence quantification model to obtain a prediction result; wherein, the event influence quantification model is obtained by training a dual-channel architecture model using historical GPU utilization data and corresponding urban governance data as a sample set, and the event influence quantification model includes a temporal convolutional network, a graph attention network, and a gated recurrent unit;

[0009] Generate a dynamic scheduling strategy based on the prediction results and the resource indicator data; wherein the dynamic scheduling strategy includes a resource allocation mechanism based on event priority, dynamically adjusts the priority of monitoring projects according to the different urban states corresponding to the urban governance data, allows high-priority events to temporarily obtain resources from low-priority events, and sets a minimum resource guarantee for each monitoring project to ensure basic operation;

[0010] Resource scheduling is performed according to the dynamic scheduling policy.

[0011] A further technical solution thereof is: after executing resource scheduling according to the dynamic scheduling strategy, the method further comprises:

[0012] Get the actual GPU usage;

[0013] The event impact weight is dynamically adjusted according to the actual GPU usage and the prediction result to adjust the event influence quantification model.

[0014] A further technical solution is as follows: the training process of the event influence quantification model includes:

[0015] Analyzing the historical GPU utilization data using a temporal convolutional network to obtain a temporal feature vector representing a cyclical trend of video memory demand;

[0016] Graph attention networks are used to analyze the relationship between specific events and resource demands corresponding to urban governance data, in order to obtain event correlation feature vectors that characterize the sudden impact of external events on resources.

[0017] The time series feature vector and the event association feature vector are fused through a gated recurrent unit, and the contribution ratio of the time series feature vector and the event association feature vector to the prediction result is dynamically adjusted.

[0018] Its further technical solution is: after obtaining the urban governance data and resource indicator data, it also includes:

[0019] The urban governance data is subjected to spatiotemporal coding processing and is processed in combination with time window slicing.

[0020] A further technical solution is: generating a dynamic scheduling strategy based on the prediction results and the resource indicator data includes:

[0021] When the prediction result exceeds a set threshold, a dynamic scheduling strategy is generated, wherein the dynamic scheduling strategy includes recovering resources from low-priority tasks or using resource slicing technology to perform resource scheduling.

[0022] A further technical solution is: the resource partitioning technology is used to perform resource scheduling, including:

[0023] The resource allocation mode is determined based on the prediction results, and resource splitting technology is used for resource scheduling. The resource allocation modes include a single-card virtualization public mode that splits the GPU into multiple instances under low load conditions for parallel use by multiple low-priority tasks, a single-card pass-through mode that occupies the entire GPU under medium load conditions, and a multi-card fusion sharing mode that merges multiple GPUs under specific conditions to support large-scale computing needs.

[0024] A further technical solution is that the dynamic scheduling strategy also includes using preset block resources when needed, and the preset block resources are kept in a standby state after initialization and are available for calling.

[0025] Its further technical solution is: the urban governance data includes traffic flow monitoring data, street crowd monitoring data, illegal parking monitoring data, fire monitoring data, engineering construction monitoring data, and activity monitoring data.

[0026] A further technical solution is: the resource indicator data includes the GPU memory occupancy rate and core utilization rate of the virtual machine.

[0027] The present invention also provides an intelligent virtualized resource dynamic scheduling method integrating urban governance data, including:

[0028] An acquisition unit, used to obtain urban governance data and resource indicator data;

[0029] A prediction unit, configured to predict a GPU demand curve for a period of time in the future based on the urban governance data input into an event influence quantification model to obtain a prediction result;

[0030] A strategy generating unit, configured to generate a dynamic scheduling strategy based on the prediction result and the resource indicator data;

[0031] An execution unit is used to execute resource scheduling according to the dynamic scheduling policy.

[0032] The beneficial effects of the present invention compared with the existing technology are: the present invention obtains comprehensive urban governance data and resource indicator data, and uses an event influence quantification model to predict the GPU demand curve in the future, thereby generating and executing a dynamic scheduling strategy based on the prediction results and real-time resource status; this method can accurately respond to specific events and high-load challenges in urban governance, ensuring that the service quality of key businesses is not affected; compared with existing resource scheduling strategies, this method emphasizes the comprehensive analysis of specific events and external data, overcoming the problem of traditional strategies being slow to respond to instantaneous changes in resource demand, thereby improving service quality and system stability, and effectively solving challenges such as untimely resource allocation and reduced service quality caused by ignoring these factors.

[0033] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0035] Figure 1 A schematic diagram of the process flow of a method for dynamic scheduling of intelligent virtualized resources integrating urban governance data provided by an embodiment of the present invention;

[0036] Figure 2 A schematic diagram of a sub-process of a method for dynamic scheduling of intelligent virtualized resources integrating urban governance data provided by an embodiment of the present invention;

[0037] Figure 3 A line chart showing the usage of the scheduling process from the start to the end of a specific event provided by an embodiment of the present invention;

[0038] Figure 4 A schematic diagram of resource preconfiguration provided in an embodiment of the present invention;

[0039] Figure 5 A schematic diagram of a flow chart of a method for dynamic scheduling of intelligent virtualized resources integrating urban governance data provided by another embodiment of the present invention;

[0040] Figure 6 A schematic block diagram of an intelligent virtualized resource dynamic scheduling system integrating urban governance data provided by an embodiment of the present invention;

[0041] Figure 7A schematic block diagram of an intelligent virtualized resource dynamic scheduling system integrating urban governance data provided by another embodiment of the present invention;

[0042] Figure 8 A schematic block diagram of a computer device provided in an embodiment of the present invention;

[0043] Figure 9 A schematic diagram of a time series database provided by an embodiment of the present invention;

[0044] Figure 10 A schematic diagram of predicting GPU memory requirements provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0046] It will be understood that when used in this specification and the appended claims, the terms “comprises” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0047] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the present invention. As used in the specification and appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0048] It should be further understood that the term "and / or" used in the present description and appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0049] See also Figure 1 , Figure 1A schematic flow chart of a method for dynamic scheduling of intelligent virtualized resources that integrates urban governance data provided in an embodiment of the present invention. The method for dynamic scheduling of intelligent virtualized resources that integrates urban governance data is applied to a server, which interacts with terminals and various virtual machines to obtain and analyze urban governance and resource indicator data, and uses a well-trained event influence quantification model to predict future GPU demand curves and generate a highly adaptable dynamic scheduling strategy. This method can not only dynamically adjust the resource allocation mode (such as single-card virtualization sharing, single-card direct pass-through, and multi-card fusion sharing) according to real-time load conditions, but also optimize model parameters based on actual usage feedback to achieve accurate response to specific events and high-load challenges, and ensure the service quality of key businesses. Compared with traditional resource scheduling strategies, the method of this embodiment effectively solves the problem of slow response caused by ignoring specific events and comprehensive analysis of external data, thereby enhancing service quality and system stability.

[0050] Figure 1 This is a flow chart of the method for dynamic scheduling of intelligent virtualized resources integrating urban governance data provided by an embodiment of the present invention. Figure 1 As shown, the method includes the following steps S110 to S140.

[0051] S110. Obtain urban governance data and resource indicator data.

[0052] In this embodiment, the urban governance data includes traffic flow monitoring data, street crowd monitoring data, illegal parking monitoring data, fire monitoring data, engineering construction monitoring data, and activity monitoring data, and the resource indicator data includes the GPU memory occupancy rate and core utilization rate of the virtual machine.

[0053] Urban governance data refers to a variety of information sources related to the city's operational status, including but not limited to traffic flow monitoring data, street crowd monitoring data, illegal parking monitoring data, fire monitoring data, construction monitoring data, and activity monitoring data. This data comes from multiple city departments and systems, such as heavy rain warnings issued by the Meteorological Bureau, water accumulation captured by traffic cameras, and environmental parameters determined by image recognition algorithms in municipal surveillance videos / cameras. Through multi-source data collection, we can gain a comprehensive understanding of the real-time status of all aspects of the city, providing a solid data foundation for predicting and responding to specific events.

[0054] Resource metric data refers to the usage of various computing resources in a virtualized environment, primarily including key performance indicators such as the virtual machine's GPU memory usage and core utilization. This data reflects the current state of the resource pool and serves as an important basis for formulating resource scheduling strategies. By monitoring resource metric data, resource bottlenecks or excesses can be promptly identified and adjusted accordingly.

[0055] Specifically, urban governance data is not limited to rainstorm warnings and waterlogging monitoring, but also includes:

[0056] Traffic flow monitoring data: used to analyze road congestion and predict the demand for video analysis and data processing during peak hours.

[0057] Street crowd monitoring data: helps assess safety risks in public places during a specific time period and identify areas that require enhanced monitoring.

[0058] Illegal parking monitoring data: helps optimize urban management efficiency and reduce the occurrence of violations.

[0059] Fire monitoring data: Provides emergency response support and ensures the rapid mobilization of necessary computing resources for data analysis.

[0060] Engineering construction monitoring data: Track dynamic changes at the construction site to ensure construction safety.

[0061] Event monitoring data: For example, crowd density predictions during events such as sports meets and concerts, to allocate appropriate resources for on-site management.

[0062] At the same time, resource metrics (such as GPU memory usage and core utilization) are crucial for maintaining efficient system operation. Combining these two types of data allows intelligent algorithms to predict future resource demands, enabling precise resource scheduling and ensuring unimpeded service quality for critical services. This approach overcomes the sluggish response of traditional resource scheduling strategies, improving overall system stability and resource utilization.

[0063] S120. Predicting the GPU demand curve for a period of time in the future based on the urban governance data input into the event influence quantification model to obtain a prediction result.

[0064] In this embodiment, prediction results are generated by inputting real-time urban governance data (such as rainstorm warnings issued by the Meteorological Bureau and the location and depth of flooded areas determined by image recognition algorithms in surveillance videos / cameras) into a trained event impact quantification model to predict virtual machine demand for GPU resources over a specific period of time. These predictions are crucial for dynamically adjusting resource allocation strategies, ensuring the timely and effective mobilization of necessary computing resources in response to specific events and ensuring the execution of critical tasks.

[0065] Specifically, the event influence quantification model is obtained by training a dual-channel architecture model using historical GPU utilization data and corresponding urban governance data as sample sets, wherein the event influence quantification model includes a temporal convolutional network, a graph attention network, and a gated recurrent unit.

[0066] In one embodiment, see Figure 2 The training process of the above-mentioned event influence quantification model includes steps S121 to S123.

[0067] S121. Analyze the historical GPU utilization data using a temporal convolutional network to obtain a temporal feature vector representing a periodic trend of video memory demand.

[0068] In this embodiment, time series feature vectors are extracted by processing historical GPU utilization data through a time series convolutional network to identify long-term trends and patterns. For example, there may be a cyclical pattern where GPU memory demand increases by 15GB between 9:00 AM and 10:00 AM on weekdays, or resource utilization is lower on weekend nights. These cyclical trends reflect resource usage patterns in daily operations and help predict baseline resource demand over the next period of time.

[0069] S122. Use the graph attention network to analyze the relationship between specific events and resource demands corresponding to urban governance data to obtain event correlation feature vectors that characterize the sudden impact of external events on resources.

[0070] In this embodiment, event-related feature vectors refer to the associations between specific events (such as rainstorm warnings, traffic flow monitoring, street crowd monitoring, illegal parking monitoring, fire monitoring, construction monitoring, and large-scale event monitoring) and corresponding resource demands, analyzed through a graph attention network. For example, when rainfall exceeds 30 mm / h, the GPU memory demand for waterlogging monitoring tasks may surge from a low daily demand (10 GB) to a high demand (60 GB). These sudden changes in resource demand are triggered by external events and therefore require special attention and quantification.

[0071] S123 , fusing the time series feature vector and the event association feature vector through a gated recurrent unit, and dynamically adjusting the contribution ratio of the time series feature vector and the event association feature vector to the prediction result.

[0072] Temporal Convolutional Network ( ) is used to analyze the long-term trend of historical GPU utilization and extract the periodic characteristics of video memory demand. For example, the video memory demand may increase by 15GB between 9:00 and 10:00 on weekdays. This part of the output is called the time series feature vector , which characterizes resource usage patterns in daily operations.

[0073] Graph Attention Network ( ): Analyze the correlation map between specific events (such as heavy rain) and related tasks (such as waterlogging monitoring) to identify the sudden impact of external events on resource demand. For example, when the rainfall exceeds 30mm / h, the GPU memory demand may surge from the daily low demand of 10GB to 60GB. This part of the output is called the event correlation feature vector , which captures the fluctuations in resource demand caused by specific events.

[0074] Gated Recurrent Unit (GRU, ): Fusion of the outputs of the two networks generates a weight α, which controls the contribution ratio of the two features (periodicity and burstiness) to the final prediction result. The formula is ,in represents the activation function (such as Sigmoid), and are the weight matrix and bias term respectively.

[0075] In this way, the event impact quantification model can accurately predict the GPU demand curve for a period of time in the future based on the current urban governance event type and severity, combined with historical data. For example, if the rainfall reaches a certain level (rainfall exceeds 30mm / h), the model predicts that the GPU memory demand will surge from the daily low demand (10GB) to the high demand (60GB), thereby triggering the corresponding resource scheduling process, such as Figure 10 shown.

[0076] This prediction mechanism is not limited to rainstorm warnings and waterlogging monitoring; it can also be extended to other urban governance events, such as traffic flow monitoring, street crowd monitoring, illegal parking monitoring, fire monitoring, construction monitoring, and large-scale event monitoring. Each event type has its own unique resource demand pattern. By learning and understanding these patterns, the system can more intelligently schedule resources, ensuring efficient operation in all circumstances. Furthermore, by continuously revising model parameters (based on the difference between actual usage and predicted values), prediction accuracy can be further improved, reducing resource waste or shortages.

[0077] S130: Generate a dynamic scheduling strategy based on the prediction result and the resource indicator data.

[0078] In this embodiment, when the prediction result exceeds a set threshold, a dynamic scheduling strategy is generated, wherein the dynamic scheduling strategy includes reclaiming resources from low-priority tasks or using resource slicing technology to perform resource scheduling.

[0079] Specifically, the resource allocation mode is determined according to the prediction results, and resource splitting technology is used for resource scheduling, wherein the resource allocation mode includes a single-card virtualization public mode that splits the GPU into multiple instances for parallel use of multiple low-priority tasks under low load conditions, a single-card pass-through mode that occupies the entire GPU under medium load conditions, and a multi-card fusion sharing mode that merges multiple GPUs under specific conditions to support large-scale computing needs.

[0080] The dynamic scheduling strategy also includes using preset block resources when needed, and the preset block resources are kept in a standby state after initialization and are available for call.

[0081] In this embodiment, when the predicted demand exceeds 100% of available resources, the system automatically reclaims resources from lower-priority tasks. For example, non-critical tasks such as historical data archiving are temporarily paused or downgraded to free up more GPU memory for higher-priority tasks.

[0082] To maximize GPU resource utilization under low load conditions, a single graphics card can be split into multiple instances (e.g., 10G, 20G, or 40G), allowing different low-priority tasks to run in parallel. This mode is called "single-card virtualization common mode."

[0083] Furthermore, the aforementioned single-GPU virtualization public mode is suitable for daily low-load scenarios (e.g., rainfall less than 10mm / h). In this case, a single GPU with 80GB of video memory is partitioned into multiple instances, each serving low-priority tasks such as historical data analysis and traffic flow monitoring. This improves resource utilization and reduces idle resources.

[0084] Single-GPU passthrough mode: In moderately intense rainstorms (e.g., rainfall rates of 30-50 mm / h), critical tasks such as video analysis of flooded areas must have sufficient computing resources. Therefore, the system uses a single-GPU passthrough mode, allowing these tasks to exclusively utilize the entire GPU's 80GB resources, eliminating performance losses associated with virtualization and ensuring real-time analysis frame rates.

[0085] Multi-GPU fusion sharing mode: In extreme disaster scenarios (rainfall > 50mm / h), facing large-scale model computing needs (such as city-wide waterlogging monitoring and prediction simulation), the system will break through the single-GPU computing bottleneck by merging three 80GB GPUs into a 240GB video memory pool to support larger-scale data processing and computing tasks.

[0086] To accelerate resource scheduling and switching, especially in scenarios requiring rapid elastic scaling or state recovery, the system uses pre-configured, fixed-configuration resource blocks. These resource blocks are pre-created and initialized, maintaining a "running" state. This enables startup times significantly faster than cold starts, enabling the creation and application of new instances in seconds or even milliseconds. This not only improves resource scheduling efficiency but also reduces resource fragmentation, increasing overall resource utilization.

[0087] Suppose the weather bureau issues a red rainstorm warning, predicting rainfall of 60 mm / h over the next few hours. The system also predicts that the GPU memory demand for the flood monitoring system will surge from the usual 10 GB to 60 GB. At this point, the system detects that the available memory in the current resource pool is only 40 GB, insufficient to meet the increased demand. Therefore, the system initiates the following scheduling strategy:

[0088] Pre-scheduling phase (T-2 hours): The system dynamically allocates or reclaims low-priority task resources from idle instances (such as temporary release for historical data training) and prepares to activate multi-card fusion mode.

[0089] Task execution phase (T0): The waterlogging monitoring service is allocated 60 GB (virtualized resources), and the GPU is simultaneously bound to the CPU core on the same node, enabling rapid resource configuration through lossless migration within seconds.

[0090] Dynamic adjustment phase (Tn): As rainfall further increases to 70 mm / h, causing demand to rise to 70 GB, the scheduler dynamically allocates additional 10 GB of fusion pool resources.

[0091] Resource recovery phase (Tn+1 hour): When rainfall weakens and drops below the safety threshold, resources are released in reverse order of business priority (first releasing 10GB of temporarily allocated resources, then automatically adjusting the 60GB fusion pool to 10GB for daily use). Fragmented video memory is merged through a dynamic reorganization algorithm and reallocated to other urban governance services, such as traffic flow analysis and personnel safety analysis.

[0092] Through the above steps, the method of this embodiment can flexibly adjust the resource allocation strategy under different scenarios, thereby ensuring the efficient operation of key tasks and achieving maximum utilization of resources.

[0093] In one embodiment, the dynamic scheduling strategy includes a resource allocation mechanism based on event priority, which dynamically adjusts the priority of monitoring projects according to different urban states corresponding to the urban governance data, allows high-priority events to temporarily obtain resources from low-priority events, and sets a minimum resource guarantee amount for each monitoring project to ensure basic operation.

[0094] Specifically, resources are dynamically allocated to respond to emergencies by adjusting event priorities and ensuring public safety and the continued operation of city functions.

[0095] The system categorizes all monitored events into four levels (P1 to P4) based on their importance and urgency. P1 is the lowest priority, primarily used for routine optimization events; P2 is used for public welfare services events; P3 targets core urban function events; and P4 is the highest priority, specifically handling events directly related to urban safety. The priority order is P1 < P2 < P3 < P4, meaning that higher-priority events can access resources from lower-priority events when needed.

[0096] In daily operations without emergencies, the priorities of various monitoring projects are as follows: water accumulation monitoring and traffic flow monitoring are rated as P2, street pedestrian flow monitoring, illegal parking monitoring, construction monitoring and activity monitoring are listed as P1, indicating that these projects are relatively less urgent or critical; fire monitoring is set as P3 due to its importance.

[0097] During heavy rainstorms, waterlogging can quickly become a major threat to public safety. Therefore, waterlogging monitoring is temporarily elevated to P4, the highest priority task. At this point, the system begins allocating resources from lower-priority tasks (such as P1 illegal parking monitoring and activity monitoring), and, if necessary, requests resources from other P2 or even P3 monitoring tasks.

[0098] During extreme heat events, fire prevention becomes the most pressing task, so fire monitoring is elevated to P4 priority. Furthermore, given the potential for high temperatures to lead to mass safety incidents during gatherings, event monitoring is also elevated to P2 priority. However, in this scenario, fire monitoring primarily draws on P1 resources to minimize impact on the more prioritized P2 event monitoring resources.

[0099] To ensure that each monitoring project can maintain basic operations even when resources are heavily occupied by emergency situations, the system configures a minimum guarantee for each monitoring item. This ensures that even under extreme conditions, necessary monitoring and data collection can continue, thereby maintaining the basic operation and safety of the city.

[0100] In summary, the system has designed a flexible and effective resource management and priority adjustment mechanism to maximize public safety while maintaining the normal operation of various urban functions.

[0101] S140: Execute resource scheduling according to the dynamic scheduling strategy.

[0102] In this embodiment, based on the prediction results, the system allocates appropriate resources (such as GPU memory) from the virtualized resource pool to meet the upcoming task requirements. These resources are temporarily placed in the expansion layer, awaiting further allocation to specific urban governance tasks.

[0103] Based on current demand and task priorities, the system prioritizes allocated resources to the most critical tasks (such as waterlogging monitoring during heavy rain).

[0104] To ensure that critical tasks have sufficient resources, resource allocations to low-priority tasks may be temporarily reduced or resources may be reclaimed from them.

[0105] If excess or shortage of resources is detected, the system will immediately take measures to adjust.

[0106] Based on the latest demand changes, the system dynamically adjusts the resource allocation strategy, which actually triggers another round of scheduling cycle.

[0107] Once the demand for an event ends, the system releases the no longer needed resources. The released resources are first returned to the expansion layer and then gradually returned to the total resource pool for use by other tasks.

[0108] Record actual data and actual usage to correct subsequent event-related feature vectors: The system records the actual data and resource usage of each event and corrects the subsequent event-related feature vectors accordingly to improve the prediction accuracy of future resource demand.

[0109] The method in this embodiment not only effectively addresses resource demand fluctuations caused by various urban governance events, but also continuously optimizes its forecasting model and resource allocation strategy, thereby improving overall resource utilization and service efficiency. Furthermore, this flexible scheduling mechanism supports rapid response to specific events, ensuring the smooth progress of urban governance work.

[0110] In one embodiment, after obtaining the urban governance data and resource indicator data, the method further includes:

[0111] The urban governance data is subjected to spatiotemporal coding processing and combined with time window slicing. Thereafter, the processed data is input into the event influence quantification model to predict the GPU demand curve in the future.

[0112] Specifically, by deploying a dedicated data collection module, urban governance data from multiple sources can be obtained in real time. These data sources include but are not limited to rainstorm warning information issued by the Meteorological Bureau, and the location and depth data of waterlogging points determined by image recognition algorithms in surveillance videos / cameras. Specifically:

[0113] Heavy rain warning information from the Meteorological Bureau: including rainfall (mm / h), warning level (such as blue, yellow, orange, red warning) and duration (hours).

[0114] Data determined by the image recognition algorithm in the surveillance video / camera: covers the specific location of the waterlogged area (latitude and longitude coordinates) and its current depth (meters or centimeters).

[0115] To consolidate the various types of collected data into a unified format for subsequent analysis and processing, spatiotemporal encoding is required. The location information of each flooding point (expressed in longitude and latitude) is converted into a geographic grid number. This can be accomplished using a predefined geographic grid system. For example, the entire city can be divided into several fixed-size grids (e.g., each with a side length of 1 kilometer). The grid number of the flooding point is then determined based on its longitude and latitude coordinates.

[0116] Dividing the timeline into specific time intervals (e.g., 10-minute periods) to form a series of time windows helps capture the changing trends of events over time and lays the foundation for subsequent time series analysis.

[0117] Within each time window, combined with the results of geographic location mapping, data from different sources are organized into a unified spatiotemporal feature matrix. This matrix not only contains the geographic grid numbers of waterlogging points but also records all relevant data within the corresponding time window (such as rainfall, warning level, and waterlogging depth). For example, if waterlogging occurs within a grid within a certain time window, the corresponding row will contain information such as waterlogging depth, rainfall, and warning level; if there is no waterlogging, the corresponding row may only contain rainfall and warning level information.

[0118] For the method of this embodiment, let's take an example of urban waterlogging monitoring. The data acquisition module deployed on the urban governance platform obtains the rainstorm warning information issued by the Meteorological Bureau in real time, covering factors such as rainfall, warning level and duration, and at the same time receives the location and depth data of the waterlogging point determined by the image recognition algorithm in the monitoring video / camera. These data are processed by spatiotemporal coding, and the longitude and latitude of the waterlogging point are mapped to the geographic grid number, combined with the time window slice with a period of 10 minutes to generate a unified spatiotemporal feature matrix. In addition, the virtualization monitoring agent, that is, the server, collects resource indicators such as GPU memory occupancy and core utilization of each virtual machine to form a resource feature vector and construct a time series database, such as Figure 9 shown.

[0119] Input the data into the model, the temporal convolutional network ( ) Analyze the long-term trend of historical GPU utilization and output a time series feature vector that represents the cyclical trend of video memory demand For example, the memory requirement increases by 15GB between 9:00-10:00 on weekdays. ) Analyze the correlation map between rainstorm events and waterlogging monitoring tasks, such as the rainstorm-waterlogging monitoring-video analysis resource demand chain, and output the event correlation feature vector reflecting the sudden impact of external events on resources , such as a sudden increase in demand due to heavy rain. The two-channel output is passed through a gated recurrent unit (GRU, ) fusion, dynamically generate weight α to control the contribution ratio of the two types of features, , predict the future GPU demand curve, such as when the rainfall exceeds 30mm / h, The GPU memory demand is predicted to surge from the usual 10GB to 60GB, triggering the resource scheduling process.

[0120] Based on the direct association of different types of events with their corresponding resource demands, for example, the demand for waterlogging monitoring will increase during heavy rains, and the demand for traffic flow and street pedestrian monitoring will also increase accordingly during holidays or peak hours. This association does not extract features through complex semantic parsing models, but establishes a more intuitive connection. A mapping relationship library is constructed based on various types of urban governance events and their corresponding resource demands. This step not only analyzes the impact of a single event, but also considers the different impacts of multiple event types on resource demands. For example, in addition to the demand for waterlogging monitoring caused by heavy rain warnings, there is also an increase in monitoring needs during large-scale events.

[0121] When prediction results exceed 100% of available resources, GPU resources are prioritized and reclaimed from lower-priority tasks, allocating computing power through resource sharding. The system employs three virtualization approaches to allocate resources: single-GPU pass-through mode (in moderate-intensity rainstorm scenarios, critical tasks exclusively utilize the entire GPU card, eliminating virtualization losses and ensuring real-time analysis frame rates); single-GPU virtualization shared mode (in routine low-load scenarios, a single GPU is split into multiple instances for parallel use by multiple tasks); and multi-GPU fusion shared mode (in extreme disaster scenarios, multiple GPUs are combined into a large memory pool to support large-scale model calculations and overcome computing bottlenecks).

[0122] like Figure 3 As shown in the figure, the system monitors resource utilization, weather, and event status in real time and dynamically allocates resources. During low-demand periods, virtualized public mode is used. During periods of heavy rain, single-card direct access or multi-card convergence mode is used as needed. After the rainfall subsides, redundant resources are reclaimed in reverse order of service priority, and a dynamic reorganization algorithm is used to consolidate fragmented graphics memory. For example, one hour after a heavy rainstorm, 70GB of graphics memory is released to the public resource pool and redistributed to other governance services.

[0123] Specifically, during the pre-scheduling phase (T-2 hours): The Meteorological Bureau issued a red rainstorm warning, and the forecast model's output of the flooding model increased its memory requirements from the usual 10GB to 60GB. The scheduler detected that only 40GB of available memory was available in the resource pool. It immediately dynamically allocated or reclaimed resources from idle instances for low-priority tasks (for example, temporarily freeing up resources for historical data training) and prepared to activate multi-GPU fusion mode.

[0124] Task execution phase (T0): The waterlogging monitoring service is allocated 60 GB (virtualized resources), and the GPU is simultaneously bound to the CPU core on the same node, enabling rapid resource configuration through lossless migration within seconds.

[0125] Dynamic adjustment phase (Tn): New waterlogging points cause the demand to rise to 70 GB, and the scheduler dynamically allocates 10 GB of fusion pool resources.

[0126] Resource recovery phase (Tn+1 hour): After the rainfall ends, resources are released in reverse order of service priority (first releasing 10 GB of temporarily allocated resources, then automatically adjusting the 60 GB fusion pool to 10 GB for daily use). A dynamic reorganization algorithm is used to release 60 GB of fragmented video memory and allocate it to other analysis services (such as traffic flow).

[0127] like Figure 4 As shown, the scheduling process incorporates a pre-configured fixed-configuration block resource mechanism. Standardized resource units are prepared in advance, and key configuration steps are completed when the block is created. Resource blocks are in the "running" state, enabling fast startup. This allows the scheduler to efficiently match physical resources, reduce fragmentation, improve resource utilization, and simplify decision-making logic.

[0128] This "provisioning" mechanism can accelerate the resource scheduling switching process, especially in scenarios that require rapid elastic scaling or state recovery, ensuring higher resource utilization and simplifying the scheduling decision logic.

[0129] Daily resource monitoring involves monitoring and analyzing urban governance data, and making preliminary resource allocations based on the data. When certain urban governance data is triggered, resource demand is predicted based on event-related characteristics, and resources from the virtualized resource pool are allocated to the expansion layer for use. These resources are then allocated to the corresponding urban governance tasks, temporarily reducing the usage of other tasks. If resource redundancy or insufficiency occurs, scheduling and adjustment are performed again until the event demand ends, and the resources are returned to the expansion layer and then to the total resource pool. Actual data is recorded to correct subsequent event-related feature vectors. Specifically, during the pre-scheduling phase, the Meteorological Bureau issues a red rainstorm warning, and the forecast model outputs an increase in memory demand for the waterlogging model. The scheduler allocates resources based on the available memory in the resource pool and prepares to activate the multi-card fusion mode. During the task execution phase, the waterlogging monitoring service is allocated resources and binds the GPU to the CPU core on the same node, achieving configuration through lossless migration in seconds. During the dynamic adjustment phase, newly added waterlogging points increase demand, and the scheduler allocates resources from the fusion pool. During the resource recovery phase, after the rainfall ends, resources are released in reverse order of business priority. The dynamic reorganization algorithm releases fragmented memory and allocates it to other analysis services.

[0130] It can be seen that the process of preliminary resource allocation based on urban governance data is as follows: first, the resource demand of a specific urban governance event is predicted based on its associated characteristics, and then the required resources are allocated from the virtualized resource pool and placed in the extension layer for use. These resources are then allocated to the corresponding urban governance tasks, and the resource usage of other tasks is temporarily reduced when necessary; when the system detects that resources are redundant or insufficient, it will reduce or increase resources, which actually starts a new round of scheduling cycle; once the event demand ends and the scheduling feedback shows no additional demand, the resources will first return to the extension layer and then return completely to the total resource pool, while recording the actual usage data and usage to correct the subsequent event-related feature vectors, thereby completing the entire resource allocation process.

[0131] The above-mentioned intelligent virtualized resource dynamic scheduling method that integrates urban governance data obtains comprehensive urban governance data and resource indicator data, and uses an event influence quantification model to predict the GPU demand curve in the future, thereby generating and executing a dynamic scheduling strategy based on the prediction results and real-time resource status; this method can accurately respond to specific events and high-load challenges in urban governance, ensuring that the service quality of key businesses is not affected; compared with existing resource scheduling strategies, this method emphasizes the comprehensive analysis of specific events and external data, overcoming the problem of traditional strategies being slow to respond to instantaneous changes in resource demand, thereby improving service quality and system stability, and effectively solving challenges such as untimely resource allocation and reduced service quality caused by ignoring these factors.

[0132] Figure 5 This is a flow chart of a method for dynamic scheduling of intelligent virtualized resources integrating urban governance data provided by another embodiment of the present invention. Figure 5As shown, the intelligent virtualized resource dynamic scheduling method integrating urban governance data in this embodiment includes steps S210-S260. Steps S210-S240 are similar to steps S110-S140 in the above embodiment and are not repeated here. The following details the added steps S250-S260 in this embodiment.

[0133] S250. Obtain actual GPU usage.

[0134] In this embodiment, real-time monitoring and recording of GPU usage by each virtual machine or container is crucial. By deploying a monitoring agent on each compute node, the system can collect resource metrics, such as GPU memory usage and core utilization, at a granular level (e.g., per second). This data not only reflects the current system operating status but also serves as an important basis for adjusting resource allocation strategies in subsequent steps.

[0135] S260: Dynamically adjust the event impact weight according to the actual GPU usage and the prediction result to adjust the event influence quantification model.

[0136] In this embodiment, the accuracy of the current event impact quantification model can be evaluated by comparing the actual GPU usage obtained with the prediction results generated by the previous prediction model. Specifically, if the predicted GPU demand value caused by a specific urban governance event (such as a rainstorm warning) is found to deviate significantly from the actual usage (too high or too low), the impact weight of the relevant event type needs to be dynamically adjusted. This process aims to optimize the event impact quantification model to make it more accurately reflect the actual impact of different events on resource demand, thereby improving the overall resource scheduling efficiency and accuracy.

[0137] If the predicted value is higher than the actual usage, it indicates that the resource demand weight currently set for this type of event may be too high, resulting in unnecessary resource reservation and potential waste. In this case, the resource demand weight for this event type should be appropriately reduced to make future resource allocation more closely aligned with actual demand.

[0138] If the predicted value is lower than the actual usage, it means that the existing model underestimates the resource demand for this type of event, which may lead to resource shortages and affect the service quality of key businesses. In this case, it is necessary to increase the corresponding resource demand weight to ensure that higher resource demands can be met in a timely manner when similar events occur in the future.

[0139] In this way, the system can continuously learn and adapt to the ever-changing urban governance environment, gradually refining its model for quantifying event impacts and improving the speed and accuracy of responses to specific events. This, in turn, maximizes resource utilization and ensures the smooth execution of various urban governance tasks. Furthermore, this feedback adjustment mechanism allows the incorporation of the latest real-world data into the model training process, further enhancing the model's ability to predict the impact of future events.

[0140] Figure 6 This is a schematic block diagram of an intelligent virtualized resource dynamic scheduling system 300 that integrates urban governance data provided by an embodiment of the present invention. Figure 6 As shown, corresponding to the above intelligent virtualized resource dynamic scheduling method integrating urban governance data, the present invention also provides an intelligent virtualized resource dynamic scheduling system 300 integrating urban governance data. The intelligent virtualized resource dynamic scheduling system 300 integrating urban governance data includes a unit for executing the above intelligent virtualized resource dynamic scheduling method integrating urban governance data, and the system can be configured in a server. Specifically, please refer to Figure 6 The intelligent virtualized resource dynamic scheduling system 300 integrating urban governance data includes an acquisition unit 301, a prediction unit 302, a strategy generation unit 303 and an execution unit 304.

[0141] An acquisition unit 301 is used to acquire urban governance data and resource indicator data; a prediction unit 302 is used to predict the GPU demand curve in the future period based on the urban governance data input into an event influence quantification model to obtain a prediction result; wherein, the event influence quantification model is obtained by training a dual-channel architecture model using historical GPU utilization data and corresponding urban governance data as a sample set, and the event influence quantification model includes a temporal convolutional network, a graph attention network, and a gated recurrent unit; a strategy generation unit 303 is used to generate a dynamic scheduling strategy based on the prediction result and the resource indicator data; an execution unit 304 is used to execute resource scheduling according to the dynamic scheduling strategy.

[0142] In one embodiment, the integrated urban governance data further includes:

[0143] A model training unit is used to analyze the historical GPU utilization data using a temporal convolutional network to obtain a temporal feature vector that characterizes the cyclical trend of video memory demand; to use a graph attention network to parse the relationship between specific events corresponding to urban governance data and resource demand to obtain an event-related feature vector that characterizes the sudden impact of external events on resources; to fuse the temporal feature vector and the event-related feature vector through a gated recurrent unit, and to dynamically adjust the contribution ratio of the temporal feature vector and the event-related feature vector to the prediction result.

[0144] In one embodiment, the apparatus further comprises:

[0145] The preprocessing unit is used to perform spatiotemporal coding processing on the urban governance data and process it in combination with time window slicing.

[0146] In one embodiment, the strategy generation unit 303 is configured to generate a dynamic scheduling strategy when the prediction result exceeds a set threshold, wherein the dynamic scheduling strategy includes recovering resources from low-priority tasks or using resource slicing technology to perform resource scheduling.

[0147] In one embodiment, the strategy generation unit 303 is further configured to determine a resource allocation mode based on the prediction results and employ resource slicing technology for resource scheduling. The resource allocation modes include a single-card virtualization common mode that divides the GPU into multiple instances for parallel use by multiple low-priority tasks under low load conditions, a single-card pass-through mode that occupies the entire GPU under medium load conditions, and a multi-card fusion sharing mode that merges multiple GPUs under specific conditions to support large-scale computing needs. The dynamic scheduling strategy also includes using preset block resources when needed. These preset block resources are initialized and remain in a standby state for invocation.

[0148] Figure 7 This is a schematic block diagram of an intelligent virtualized resource dynamic scheduling system 300 that integrates urban governance data, provided by another embodiment of the present invention. Figure 7 As shown, the intelligent virtualized resource dynamic scheduling system 300 integrating urban governance data of this embodiment is based on the above embodiment and adds a usage acquisition unit 305 and an adjustment unit 306.

[0149] The usage acquisition unit 305 is used to obtain the actual GPU usage; the adjustment unit 306 is used to dynamically adjust the event impact weight according to the actual GPU usage and the prediction result to adjust the event influence quantification model.

[0150] It should be noted that technical personnel in the relevant field can clearly understand that the specific implementation process of the above-mentioned intelligent virtualized resource dynamic scheduling system 300 and each unit that integrates urban governance data can refer to the corresponding description in the aforementioned method embodiment. For the convenience and conciseness of the description, it will not be repeated here.

[0151] The intelligent virtualized resource dynamic scheduling system 300 integrating urban governance data can be implemented in the form of a computer program. Figure 8 Runs on the computer equipment shown.

[0152] See also Figure 8 , Figure 8 1 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device 500 may be a server, wherein the server may be an independent server or a server cluster composed of multiple servers.

[0153] See Figure 8 The computer device 500 includes a processor 502 , a memory, and a network interface 505 connected via a system bus 501 , wherein the memory may include a non-volatile storage medium 503 and an internal memory 504 .

[0154] The non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions, which, when executed, enable the processor 502 to execute a method for dynamic scheduling of intelligent virtualized resources that integrates urban governance data.

[0155] The processor 502 is used to provide computing and control capabilities to support the operation of the entire computer device 500.

[0156] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute an intelligent virtualized resource dynamic scheduling method that integrates urban governance data.

[0157] The network interface 505 is used to communicate with other devices through the network. Figure 8 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present application, and does not constitute a limitation on the computer device 500 to which the solution of the present application is applied. The specific computer device 500 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0158] The processor 502 is configured to execute a computer program 5032 stored in the memory to implement the following steps:

[0159] Obtain urban governance data and resource indicator data; predict the GPU demand curve in the future period based on the urban governance data input into the event influence quantification model to obtain a prediction result; generate a dynamic scheduling strategy based on the prediction result and the resource indicator data; and execute resource scheduling according to the dynamic scheduling strategy.

[0160] The event influence quantification model is obtained by training a dual-channel architecture model using historical GPU utilization data and corresponding urban governance data as sample sets, wherein the event influence quantification model includes a temporal convolutional network, a graph attention network, and a gated recurrent unit.

[0161] The urban governance data includes traffic flow monitoring data, street crowd monitoring data, illegal parking monitoring data, fire monitoring data, engineering construction monitoring data, and activity monitoring data.

[0162] The resource indicator data includes the GPU memory occupancy rate and core utilization rate of the virtual machine.

[0163] Among them, the dynamic scheduling strategy includes a resource allocation mechanism based on event priority, dynamically adjusts the priority of monitoring projects according to the different urban status corresponding to the urban governance data, and allows high-priority events to temporarily obtain resources from low-priority events. At the same time, a minimum resource guarantee amount is set for each monitoring project to ensure basic operation.

[0164] In one embodiment, after implementing the step of performing resource scheduling according to the dynamic scheduling policy, the processor 502 further implements the following steps:

[0165] Obtaining actual GPU usage; dynamically adjusting the event impact weight according to the actual GPU usage and the prediction result to adjust the event influence quantification model.

[0166] In one embodiment, when implementing the training step of the event influence quantification model, the processor 502 specifically implements the following steps:

[0167] The historical GPU utilization data is analyzed using a temporal convolutional network to obtain a temporal feature vector that characterizes the cyclical trend of video memory demand. The relationship between specific events and resource demand corresponding to urban governance data is analyzed using a graph attention network to obtain an event-related feature vector that characterizes the sudden impact of external events on resources. The temporal feature vector and the event-related feature vector are fused through a gated recurrent unit, and the contribution ratio of the temporal feature vector and the event-related feature vector to the prediction result is dynamically adjusted.

[0168] In one embodiment, after implementing the step of obtaining the urban governance data and resource indicator data, the processor 502 further implements the following steps:

[0169] The urban governance data is subjected to spatiotemporal coding processing and is processed in combination with time window slicing.

[0170] In one embodiment, when the processor 502 implements the step of generating a dynamic scheduling strategy according to the prediction result and the resource indicator data, the processor 502 specifically implements the following steps:

[0171] When the prediction result exceeds a set threshold, a dynamic scheduling strategy is generated, wherein the dynamic scheduling strategy includes recovering resources from low-priority tasks or using resource slicing technology to perform resource scheduling.

[0172] The dynamic scheduling strategy also includes using preset block resources when needed, and the preset block resources are kept in a standby state after initialization and are available for call.

[0173] In one embodiment, when implementing the step of using the resource segmentation technology to perform resource scheduling, the processor 502 specifically implements the following steps:

[0174] The resource allocation mode is determined based on the prediction results, and resource splitting technology is used for resource scheduling. The resource allocation modes include a single-card virtualization public mode that splits the GPU into multiple instances under low load conditions for parallel use by multiple low-priority tasks, a single-card pass-through mode that occupies the entire GPU under medium load conditions, and a multi-card fusion sharing mode that merges multiple GPUs under specific conditions to support large-scale computing needs.

[0175] It should be understood that in the embodiment of the present application, the processor 502 may be a central processing unit (CPU), and the processor 502 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0176] Those skilled in the art will appreciate that all or part of the steps in the method of the above-described embodiment can be implemented by instructing the relevant hardware through a computer program. The computer program includes program instructions, which can be stored in a storage medium that is computer-readable. The program instructions are executed by at least one processor in the computer system to implement the steps in the method of the above-described embodiment.

[0177] Therefore, the present invention also provides a storage medium. The storage medium may be a computer-readable storage medium. The storage medium stores a computer program, wherein when the computer program is executed by a processor, the processor performs the following steps:

[0178] Obtain urban governance data and resource indicator data; predict the GPU demand curve in the future period based on the urban governance data input into the event influence quantification model to obtain a prediction result; generate a dynamic scheduling strategy based on the prediction result and the resource indicator data; and execute resource scheduling according to the dynamic scheduling strategy.

[0179] The event influence quantification model is obtained by training a dual-channel architecture model using historical GPU utilization data and corresponding urban governance data as sample sets, wherein the event influence quantification model includes a temporal convolutional network, a graph attention network, and a gated recurrent unit.

[0180] The urban governance data includes traffic flow monitoring data, street crowd monitoring data, illegal parking monitoring data, fire monitoring data, engineering construction monitoring data, and activity monitoring data.

[0181] The resource indicator data includes the GPU memory occupancy rate and core utilization rate of the virtual machine.

[0182] Among them, the dynamic scheduling strategy includes a resource allocation mechanism based on event priority, dynamically adjusts the priority of monitoring projects according to the different urban status corresponding to the urban governance data, and allows high-priority events to temporarily obtain resources from low-priority events. At the same time, a minimum resource guarantee amount is set for each monitoring project to ensure basic operation.

[0183] In one embodiment, after executing the computer program to implement the step of performing resource scheduling according to the dynamic scheduling policy, the processor further implements the following steps:

[0184] Obtaining actual GPU usage; dynamically adjusting the event impact weight according to the actual GPU usage and the prediction result to adjust the event influence quantification model.

[0185] In one embodiment, when the processor executes the computer program to implement the training step of the event influence quantification model, the processor specifically implements the following steps:

[0186] The historical GPU utilization data is analyzed using a temporal convolutional network to obtain a temporal feature vector that characterizes the cyclical trend of video memory demand. The relationship between specific events and resource demand corresponding to urban governance data is analyzed using a graph attention network to obtain an event-related feature vector that characterizes the sudden impact of external events on resources. The temporal feature vector and the event-related feature vector are fused through a gated recurrent unit, and the contribution ratio of the temporal feature vector and the event-related feature vector to the prediction result is dynamically adjusted.

[0187] In one embodiment, after executing the computer program to implement the step of obtaining urban governance data and resource indicator data, the processor further implements the following steps:

[0188] The urban governance data is subjected to spatiotemporal coding processing and is processed in combination with time window slicing.

[0189] In one embodiment, when the processor executes the computer program to implement the step of generating a dynamic scheduling strategy based on the prediction result and the resource indicator data, the processor specifically implements the following steps:

[0190] When the prediction result exceeds a set threshold, a dynamic scheduling strategy is generated, wherein the dynamic scheduling strategy includes recovering resources from low-priority tasks or using resource slicing technology to perform resource scheduling.

[0191] The dynamic scheduling strategy also includes using preset block resources when needed, and the preset block resources are kept in a standby state after initialization and are available for call.

[0192] In one embodiment, when the processor executes the computer program to implement the step of using resource slicing technology to perform resource scheduling, the processor specifically implements the following steps:

[0193] The resource allocation mode is determined based on the prediction results, and resource splitting technology is used for resource scheduling. The resource allocation modes include a single-card virtualization public mode that splits the GPU into multiple instances under low load conditions for parallel use by multiple low-priority tasks, a single-card pass-through mode that occupies the entire GPU under medium load conditions, and a multi-card fusion sharing mode that merges multiple GPUs under specific conditions to support large-scale computing needs.

[0194] The storage medium may be any computer-readable storage medium that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk.

[0195] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the composition and steps of each example according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0196] In the several embodiments provided herein, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of the various units is merely a logical functional division, and actual implementations may employ other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be omitted or not implemented.

[0197] The steps in the method of the embodiment of the present invention may be adjusted in order, combined, or deleted as needed. The units in the system of the embodiment of the present invention may be combined, divided, or deleted as needed. In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit.

[0198] If this integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes instructions for enabling a computer device (such as a personal computer, terminal, or network device) to execute all or part of the steps of the method described in various embodiments of the present invention.

[0199] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions are intended to be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. An intelligent virtualized resource dynamic scheduling method integrating urban governance data, characterized by: include: Obtain urban governance data and resource indicator data; The GPU demand curve for a period of time in the future is predicted based on the input of the urban governance data into the event influence quantification model to obtain a prediction result; wherein, the event influence quantification model is obtained by training a dual-channel architecture model using historical GPU utilization data and corresponding urban governance data as a sample set, and the event influence quantification model includes a temporal convolutional network, a graph attention network, and a gated recurrent unit; Generate a dynamic scheduling strategy based on the prediction results and the resource indicator data; wherein the dynamic scheduling strategy includes a resource allocation mechanism based on event priority, dynamically adjusts the priority of monitoring projects according to the different urban states corresponding to the urban governance data, allows high-priority events to temporarily obtain resources from low-priority events, and sets a minimum resource guarantee for each monitoring project to ensure basic operation; Resource scheduling is performed according to the dynamic scheduling policy.

2. The intelligent virtualized resource dynamic scheduling method integrating urban governance data according to claim 1 is characterized in that: After executing resource scheduling according to the dynamic scheduling strategy, the method further includes: Get the actual GPU usage; The event impact weight is dynamically adjusted according to the actual GPU usage and the prediction result to adjust the event influence quantification model.

3. The intelligent virtualized resource dynamic scheduling method integrating urban governance data according to claim 1 is characterized in that: The training process of the event influence quantification model includes: Analyzing the historical GPU utilization data using a temporal convolutional network to obtain a temporal feature vector representing a cyclical trend of video memory demand; Graph attention networks are used to analyze the relationship between specific events corresponding to urban governance data and resource demands to obtain event correlation feature vectors that characterize the sudden impact of external events on resources. These specific events include rainstorm warnings, traffic flow monitoring, street crowd monitoring, illegal parking monitoring, fire monitoring, construction monitoring, and large-scale event monitoring. The time series feature vector and the event association feature vector are fused through a gated recurrent unit, and the contribution ratio of the time series feature vector and the event association feature vector to the prediction result is dynamically adjusted.

4. The intelligent virtualized resource dynamic scheduling method integrating urban governance data according to claim 1 is characterized in that: After obtaining the urban governance data and resource indicator data, the following steps are also included: The urban governance data is subjected to spatiotemporal coding processing and is processed in combination with time window slicing.

5. The intelligent virtualized resource dynamic scheduling method integrating urban governance data according to claim 1 is characterized in that: Generating a dynamic scheduling strategy according to the prediction result and the resource indicator data includes: When the prediction result exceeds a set threshold, a dynamic scheduling strategy is generated, wherein the dynamic scheduling strategy includes recovering resources from low-priority tasks or using resource slicing technology to perform resource scheduling.

6. The intelligent virtualized resource dynamic scheduling method integrating urban governance data according to claim 5 is characterized in that: The resource partitioning technology is used to perform resource scheduling, including: The resource allocation mode is determined based on the prediction results, and resource splitting technology is used for resource scheduling. The resource allocation modes include a single-card virtualization public mode that splits the GPU into multiple instances for parallel use by multiple low-priority tasks under low load conditions, a single-card pass-through mode that occupies the entire GPU under medium load conditions, and a multi-card fusion sharing mode that merges multiple GPUs to support large-scale computing needs in extreme disaster scenarios.

7. The intelligent virtualized resource dynamic scheduling method integrating urban governance data according to claim 5 is characterized in that: The dynamic scheduling strategy also includes using preset block resources when needed, and the preset block resources are kept in a standby state after initialization and are available for call.

8. The intelligent virtualized resource dynamic scheduling method integrating urban governance data according to claim 1 is characterized in that: The urban governance data includes traffic flow monitoring data, street crowd monitoring data, illegal parking monitoring data, fire monitoring data, engineering construction monitoring data, and activity monitoring data.

9. The intelligent virtualized resource dynamic scheduling method integrating urban governance data according to claim 1 is characterized in that: The resource indicator data includes the GPU memory occupancy rate and core utilization rate of the virtual machine.

10. An intelligent virtualized resource dynamic scheduling system integrating urban governance data is characterized by: The system uses the intelligent virtualized resource dynamic scheduling method integrating urban governance data as described in claim 1, including: An acquisition unit, used to obtain urban governance data and resource indicator data; A prediction unit, configured to predict a GPU demand curve for a period of time in the future based on the urban governance data input into an event influence quantification model to obtain a prediction result; A strategy generating unit, configured to generate a dynamic scheduling strategy based on the prediction result and the resource indicator data; An execution unit is used to execute resource scheduling according to the dynamic scheduling policy.

Images