Business-aware-based carbon satellite ground system resource dynamic allocation method

Abstract

The application provides a service-aware-based dynamic allocation method for carbon satellite ground system resources, relates to the technical field of computing power resource allocation, and comprises the following steps: through analysis, a benchmark task priority distribution of a benchmark task node distribution is obtained, and then a container resource pre-call distribution is output after execution container scheduling timing analysis; a high-load node sequence is projected to the container resource pre-call distribution, and a container resource pre-call sequence is located; after judging the load demand of real-time weather satellite and carbon satellite data streams, the container resource pre-call sequence is triggered to perform timing elastic scheduling of container resources. The application solves the defects of the prior art, such as the rigidity of ground system resource allocation, low utilization rate of task processing container resources, and serious delay of carbon satellite monitoring data processing. The application achieves the technical effects of multi-dimensional resource pre-scheduling strategy and elastic feedback closed-loop control mechanism, realizes dynamic load-driven efficient resource matching, and significantly improves the utilization rate of heterogeneous container resources.

Description

Technical Field

[0001] This invention relates to the field of computing resource allocation technology, and in particular to a method for dynamic resource allocation of carbon satellite ground systems based on business awareness. Background Technology

[0002] Traditional satellite ground system resource allocation mechanisms mostly adopt static preset strategies, configuring container resources based on fixed time windows or average load estimates.

[0003] Such solutions are ill-suited to the spatiotemporal clustering effect of loads caused by the cyclical nature and geographical correlation of industrial activities in carbon satellite monitoring missions, resulting in two typical problems: First, redundant container resources are idle for a long time during normal periods, and the utilization rate remains low; Second, there is insufficient supply of spatiotemporal node resources during high load periods, and emergency monitoring missions enter a queue due to container preemption failures, leading to a deterioration in the timeliness of critical data processing.

[0004] Especially in scenarios where the Fengyun satellite ground system and the carbon satellite share resources, the carbon satellite is constrained by the resources of scientific research projects and urgently needs to reuse the idle container resources of the Fengyun system. However, existing technologies cannot achieve secure resource sharing while ensuring the stability and timeliness of Fengyun satellite services: if resources are statically reserved for the carbon satellite, it will exacerbate the idle resources of the Fengyun system; if the carbon satellite is allowed to seize resources, it may lead to insufficient container resources for Fengyun services and deterioration in timeliness.

[0005] In summary, existing technologies suffer from rigid allocation of ground system resources, resulting in low utilization of mission processing container resources and severe delays in carbon satellite monitoring data processing. Alternatively, the carbon satellite mission processing container may consume excessive resources, preventing timely processing of Fengyun satellite mission data. Summary of the Invention

[0006] This invention provides a service-aware method for dynamic allocation of ground system resources for carbon satellites. It addresses the technical problems of rigid allocation of ground system resources in existing technologies, which leads to low utilization of mission processing container resources and severe delays in carbon satellite monitoring data processing, or the problem of carbon satellite mission processing containers occupying too many resources, resulting in the inability to process Fengyun satellite missions in a timely manner.

[0007] In view of the above problems, the present invention provides a method for dynamic allocation of carbon satellite ground system resources based on business awareness. The method includes: performing load spatiotemporal correlation analysis on historical mission data of Fengyun satellites accessed from the cloud, and outputting a load spatiotemporal distribution cloud map; identifying cross-regional load transmission on the load spatiotemporal distribution cloud map, and outputting a high-load spatiotemporal node distribution; locating the baseline mission node distribution of the high-load spatiotemporal node distribution through spatiotemporal projection matching; collecting mission feature vectors from the baseline mission node distribution, and performing mission priority analysis based on vector labels to output a baseline mission priority distribution; performing container scheduling timing analysis on the baseline mission node distribution with the baseline mission priority distribution as a constraint, and outputting a container resource pre-call distribution; after defining the high-load node sequence in the high-load spatiotemporal node distribution based on the container call time window, projecting the high-load node sequence onto the container resource pre-call distribution to locate the container resource pre-call sequence; and triggering the container resource pre-call sequence for time-series elastic scheduling of container resources after judging the load demand of the real-time carbon satellite data stream.

[0008] In one implementation, load spatiotemporal correlation analysis is performed on historical mission data of Fengyun satellites accessed from the cloud to output a load spatiotemporal distribution cloud map, and the following processing is also performed:

[0009] Based on the storage years, historical mission data of Fengyun satellites are decomposed to obtain multi-year carbon satellite mission data. Spatiotemporal data modeling is performed on the multi-year carbon satellite mission data to obtain multiple three-dimensional load matrices. Based on the load intensity visualization mapping rules, the load intensity mapping of the multiple three-dimensional load matrices is converted into color gradients to generate multiple historical load spatiotemporal distributions. After time-series alignment and superposition of the multiple historical load spatiotemporal distributions, the periodic load transmission pattern is mined through spatiotemporal correlation analysis, and the load spatiotemporal distribution cloud map is output.

[0010] In one implementation, spatiotemporal data modeling is performed on the multi-year carbon satellite mission data to obtain multiple three-dimensional payload matrices, and the following processing is also performed:

[0011] The first-year carbon satellite mission data is decomposed into attributes to obtain multiple mission attribute information; a spatiotemporal grid model is constructed using the index composition of the mission attribute information as the axis; the multiple mission attribute information is filled into the spatiotemporal grid model to obtain a first three-dimensional load matrix; after decomposing the multi-year carbon satellite mission data, the spatiotemporal grid model is reused and filled to obtain the multiple three-dimensional load matrices.

[0012] In one implementation, the load spatiotemporal distribution cloud map is subjected to cross-regional load conduction identification, and the distribution of high-load spatiotemporal nodes is output. The following processing is also performed:

[0013] Based on the industrial distribution characteristics of carbon emissions, the spatiotemporal distribution cloud map of the load is divided into multiple key observation areas, and a spatial adjacency matrix between these key observation areas is constructed. A temporal correlation analysis of load intensity between the key observation areas is performed along the spatial adjacency matrix to obtain the cross-regional load transmission path topology, where the out-degree topology connection line is marked with a transmission delay. A predefined load intensity threshold is used to traverse the spatiotemporal distribution cloud map of the load to locate candidate high-load node distributions. Based on the cross-regional load transmission path topology, the candidate high-load node distributions are compensated for by transmission effects to identify additional high-load node distributions caused by cross-regional transmission. The candidate high-load node distributions and the additional high-load node distributions are spatiotemporally aligned to output the high-load spatiotemporal node distribution.

[0014] In one implementation, after collecting task feature vectors from the baseline task node distribution, task priority analysis is performed based on the vector labels to output the baseline task priority distribution. The following processing is also performed:

[0015] Extract multiple first task feature vectors from multiple first benchmark tasks in the first benchmark task node; normalize the multiple first task features to obtain multiple first task value scores; serialize the multiple first benchmark tasks according to the multiple first task value scores to obtain a first initial priority; dynamically correct the first initial priority according to the task dependency chain of the multiple first benchmark tasks, and output the first benchmark priority; and so on, after collecting task feature vectors from the distribution of the benchmark task nodes, perform task priority analysis based on vector labels to output the benchmark task priority distribution.

[0016] In one implementation, the task feature vector is composed of task timeliness, data value, resource consumption, and processing time.

[0017] In one implementation, using the baseline task priority distribution as a constraint, container scheduling timing analysis is performed on the baseline task node distribution to output the container resource pre-call distribution, and the following processing is also performed:

[0018] The system invokes multiple first resource requirement vectors from the multiple first benchmark tasks, wherein the resource requirement vectors include container capacity requirements, computing power requirements, memory requirements, and I / O bandwidth requirements; heterogeneous container matching is performed based on the multiple first resource requirement vectors to obtain multiple heterogeneous container codes; container scheduling timing deployment is performed based on code consistency and the first benchmark priority to output the first container resource pre-call rule; and so on, with the benchmark task priority distribution as a constraint, container scheduling timing analysis is performed on the benchmark task node distribution to output the container resource pre-call distribution.

[0019] In one implementation, the task attribute information includes task time, data stream source space, and quantified task load.

[0020] The technical solution provided in this invention has at least the following technical effects or advantages:

[0021] The method provided in this invention analyzes the spatiotemporal load correlation characteristics of historical mission data from Fengyun satellites to construct a multi-dimensional spatiotemporal load distribution cloud map, and identifies high-load spatiotemporal nodes based on cross-regional load transmission patterns. It then locates the historical baseline task set corresponding to the high-load nodes through a spatiotemporal mapping matching mechanism, and generates a baseline task priority distribution by combining task feature vectors and a dynamic priority evaluation model. Based on this, it optimizes the timing of containerized resource scheduling using the priority distribution as a constraint, generates a pre-call strategy matrix, and triggers an elastic scaling scheduling mechanism through dynamic matching of container resource pre-call sequences with real-time load demands, enabling precise supply and utilization of computing resources. This achieves the technical effect of realizing efficient resource matching driven by dynamic load based on a multi-dimensional resource pre-scheduling strategy and an elastic feedback closed-loop control mechanism, significantly improving the utilization rate of heterogeneous container resources. Attached Figure Description

[0022] Figure 1 This invention illustrates a flowchart of a service-aware dynamic resource allocation method for carbon satellite ground systems provided by the present invention.

[0023] Figure 2 This paper illustrates a flowchart of the process for locating the distribution of high-load spatiotemporal nodes in the operationally aware dynamic resource allocation method for carbon satellite ground systems provided by the present invention. Detailed Implementation

[0024] This invention provides a service-aware method for dynamic allocation of ground system resources for carbon satellites. It addresses the technical problems of rigid allocation of ground system resources in existing technologies, which leads to low utilization of mission processing container resources and severe delays in carbon satellite monitoring data processing, or the problem of carbon satellite mission processing containers occupying too many resources, resulting in the inability to process Fengyun satellite missions in a timely manner.

[0025] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. It should also be noted that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, not all of them.

[0026] The flowchart of the service-aware dynamic resource allocation method for carbon satellite ground systems provided in this embodiment of the invention is shown below. Figure 1 The method includes:

[0027] Step A100: Perform load spatiotemporal correlation analysis on the historical mission data of Fengyun satellites retrieved from the cloud and output a load spatiotemporal distribution cloud map.

[0028] In one implementation, load spatiotemporal correlation analysis is performed on historical mission data of Fengyun satellites retrieved from the cloud, and a load spatiotemporal distribution cloud map is output. Step A100 of the method includes:

[0029] Step A110: Decompose the historical mission data of Fengyun satellites based on the storage years to obtain multi-year carbon satellite mission data.

[0030] Step A120: Perform spatiotemporal data modeling on the multi-year carbon satellite mission data to obtain multiple three-dimensional load matrices.

[0031] Step A130: Based on the load intensity visualization mapping rules, convert the load intensity mapping of the multiple three-dimensional load matrices into color gradients to generate multiple historical load spatiotemporal distributions.

[0032] Step A140: After time-series alignment and overlay of the multiple historical load spatiotemporal distributions, the periodic load transmission pattern is mined through spatiotemporal correlation analysis, and the load spatiotemporal distribution cloud map is output.

[0033] In one embodiment, spatiotemporal data modeling is performed on the multi-year carbon satellite mission data to obtain multiple three-dimensional payload matrices, and step A120 of the method includes:

[0034] Step A121: Perform attribute decomposition on the first year's carbon satellite mission data to obtain multiple mission attribute information.

[0035] Step A122: Construct a spatiotemporal grid model using the index composition of the task attribute information as the axis.

[0036] Step A123: Fill the spatiotemporal grid model with the multiple task attribute information to obtain the first three-dimensional load matrix.

[0037] Step A124: After performing attribute decomposition on the multi-year carbon satellite mission data, the spatiotemporal grid model is reused and filled to obtain the multiple three-dimensional load matrices.

[0038] In one implementation, the task attribute information includes task time, data stream source space, and quantified task load.

[0039] This embodiment calls upon historical mission data of Fengyun satellites stored in the cloud to analyze the load distribution patterns in different time and spatial dimensions, generating a visualized load spatiotemporal distribution cloud map. The load spatiotemporal distribution cloud map intuitively displays the load intensity with color gradients, helping to identify high-load areas and time windows, and providing data support for subsequent resource scheduling.

[0040] Specifically, this implementation decomposes the historical mission data of the Fengyun satellites according to their storage years, such as by year, to separate mission datasets from different years and obtain multi-year carbon satellite mission data. This decomposition method facilitates the analysis of long-term trends in payload evolution over time, while avoiding interference from patterns caused by data from different years being mixed together.

[0041] Key attributes are extracted from task data of a single year to obtain multiple task attribute information. Each task attribute information specifically includes task execution time, geographical location of data source in latitude and longitude form, and quantitative load indicators such as CPU / memory consumption. These attributes provide basic data units for spatiotemporal modeling.

[0042] The spatiotemporal grid model is constructed with time, spatial coordinates, and load intensity as three orthogonal axes. The time axis of the spatiotemporal grid model is sliced ​​at fixed intervals, such as 15-minute slices, the spatial axis is divided according to geographic grids, such as 1°×1° latitude and longitude, and the load axis quantifies the intensity of task resource consumption.

[0043] The decomposed task attribute data is filled into the grid model according to spatiotemporal coordinates to generate the first three-dimensional load matrix for the first observation year. The value of each cell in the first three-dimensional load matrix represents the average load intensity of that time slice and geographical region, forming a digital representation of spatiotemporal load.

[0044] Similarly, by reusing the same spatiotemporal grid model structure, the remaining carbon satellite mission data from the multi-year carbon satellite mission data are filled in according to the same rules to generate the multiple three-dimensional load matrices. This reuse in this embodiment ensures the spatiotemporal alignment of data from different years, facilitating subsequent overlay analysis and pattern mining.

[0045] Based on load intensity visualization mapping rules, such as blue for low load and red for high load, load intensity values ​​in multiple three-dimensional load matrices are mapped to color gradients to generate historical load spatiotemporal distribution maps for each year. These historical load spatiotemporal distribution maps visually display the spatiotemporal clustering characteristics of loads in the form of heat maps.

[0046] By aligning and overlaying historical load spatiotemporal distribution maps from multiple years along the time axis, we can analyze cross-year patterns such as cyclical loads with high loads at fixed times each day, and spatial transmission paths such as load surges in one region triggering subsequent load increases in adjacent regions. The resulting load spatiotemporal distribution cloud map integrates long-term patterns, providing a basis for predicting future loads.

[0047] Step A200: Perform cross-regional load conduction identification on the load spatiotemporal distribution cloud map and output the high load spatiotemporal node distribution.

[0048] In one implementation, such as Figure 2 As shown, the method performs cross-regional load conduction identification on the load spatiotemporal distribution cloud map and outputs the high-load spatiotemporal node distribution. Step A200 of the method includes:

[0049] Step A210: Based on the industrial distribution characteristics of carbon emissions, the spatiotemporal distribution cloud map of the load is divided into multiple key observation areas, and a spatial adjacency matrix between the multiple key observation areas is constructed.

[0050] Step A220: Perform inter-regional load intensity temporal correlation analysis on the multiple key observation areas along the spatial adjacency relation matrix to obtain the cross-regional load transmission path topology, wherein the transmission delay is marked on the out-degree topology connection line.

[0051] Step A230: Predefine a load intensity threshold, and use the load intensity threshold to traverse the load spatiotemporal distribution cloud map to locate the distribution of candidate high-load nodes.

[0052] Step A240: Based on the cross-regional load transmission path topology, perform transmission effect compensation correction on the candidate high-load node distribution to identify the additional high-load node distribution caused by cross-regional transmission.

[0053] Step A250: Spatiotemporally align the candidate high-load node distribution and the additional high-load node distribution, and output the high-load spatiotemporal node distribution.

[0054] Specifically, this embodiment uses data such as the coordinates of thermal power plants and the industrial distribution data of carbon emissions in industrial clusters as a benchmark to divide the geographic space of the cloud map into multiple key observation areas. Then, based on geographic boundary proximity, such as adjacent latitude and longitude or connected transportation networks, an adjacency matrix is ​​constructed between these multiple key observation areas, marking directly connected pairs of areas to provide a spatial correlation framework for subsequent transmission path analysis.

[0055] A time-series correlation analysis of load intensity is performed on adjacent regions. For example, using a lag cross-correlation algorithm, the correlation between the load intensity of region A at time t and the load intensity of region B at time t+Δt is calculated. If the correlation is significant and there is a time lag, such as Δt=2 hours, then a load transmission path from region A to region B is determined to exist. The transmission path topology graph uses directed edges to represent the transmission direction and edge weights to indicate the transmission delay.

[0056] The mean and standard deviation of load intensity are calculated based on historical load data. The load intensity threshold, such as μ+2σ, is dynamically set. The spatiotemporal cells in the cloud map are traversed, and cells with load intensity exceeding the threshold are marked as candidate high-load nodes. For example, if the load intensity of a certain region continuously exceeds the threshold from 08:00 to 10:00 UTC, it is marked as a candidate node. Finally, the distribution of the candidate high-load nodes is located and output.

[0057] Based on the cross-regional load transmission path topology, additional nodes that were not directly detected but were caused by transmission are identified among the candidate nodes. For example, if a node exceeds the load threshold between 10:00 and 12:00 UTC, and its time window matches the transmission delay of the upstream region, then the upstream node is determined to be an additional high-load node caused by cross-regional transmission.

[0058] Similarly, based on the cross-regional load transmission path topology, the distribution of candidate high-load nodes is compensated and corrected for transmission effects to identify the additional high-load node distribution caused by cross-regional transmission.

[0059] The original candidate nodes (i.e., candidate high-load nodes) and the transmission-related nodes (i.e., additional high-load nodes) are aligned according to their spatiotemporal coordinates. Duplicates are merged and conflicts are resolved. For example, a conflict here could be that the same spatiotemporal unit is marked multiple times, generating the final spatiotemporal distribution map of high-load nodes. Each node in the map is labeled with its spatiotemporal boundary, load intensity level, and associated transmission path, such as UTC 08:00-10:00, longitude 115°-120°, transmission originating from industrial area A.

[0060] This embodiment aims to analyze the transmission effect of load across regions, identify high-load spatiotemporal nodes caused by regional correlation, and ultimately generate a distribution of high-load spatiotemporal nodes covering both native and transmission types, providing a precise location basis for subsequent container-level resource scheduling and cross-container resource scheduling.

[0061] Step A300: Locate the baseline task node distribution of the high-load spatiotemporal node distribution by spatiotemporal projection matching.

[0062] Specifically, in this embodiment, based on the high-load spatiotemporal node distribution generated in step A200, its spatiotemporal coordinates are mapped inversely to the historical task database, and a set of historical tasks that match the spatiotemporal attributes of the current prediction node is retrieved. Here, the spatiotemporal coordinates include time windows and geographical ranges.

[0063] For example, if the predicted node is Industrial Zone A between 08:00 and 10:00 UTC on March 30, 2025, then the set of tasks executed in that area during the same period in the past is extracted from historical data to form the baseline task node distribution.

[0064] This process establishes a correlation between the predicted scenario and historical experience through spatiotemporal projection, providing a comparable set of task samples for priority assessment.

[0065] Step A400: After collecting task feature vectors from the baseline task node distribution, perform task priority analysis based on the vector labels and output the baseline task priority distribution.

[0066] In one implementation, after collecting task feature vectors from the baseline task node distribution, task priority analysis is performed based on the vector labels to output the baseline task priority distribution. Step A400 of the method includes:

[0067] Step A410: Extract multiple first task feature vectors from multiple first benchmark tasks in the first benchmark task node.

[0068] Step A420: Normalize the multiple first task features to obtain multiple first task value scores.

[0069] Step A430: Serialize the multiple first benchmark tasks according to the multiple first task value scores to obtain the first initial priority.

[0070] Step A440: Based on the task dependency chain of the plurality of first baseline tasks, dynamically correct the first initial priority by priority increment, and output the first baseline priority.

[0071] Step A450: Similarly, after collecting task feature vectors from the baseline task node distribution, perform task priority analysis based on the vector labels and output the baseline task priority distribution.

[0072] In one implementation, the task feature vector is composed of task timeliness, data value, resource consumption, and processing time.

[0073] Specifically, in this embodiment, the first benchmark task node is any node in the distribution of benchmark task nodes, and since the data processing logic of each node is consistent, this embodiment takes the data processing process of the first benchmark task node as an example to provide a concise explanation of the technical solution.

[0074] Calculate the timeliness indicators (such as the reciprocal of the remaining time), data value weights (assigned according to the industrial level of the monitoring area), resource consumption coefficients (standardized CPU / memory requirements), and processing time (normalized to 0-1 values) of multiple first benchmark tasks in the first benchmark task node to construct multiple structured feature vectors, thereby obtaining the multiple first task feature vectors.

[0075] For example, the timeliness index of the benchmark task can be the reciprocal of the remaining time, the data value weight can be assigned according to the industrial level of the monitored area, the resource consumption coefficient can be the standardized CPU / memory requirement, and the processing time is preferably normalized to a value of 0-1.

[0076] The characteristics of the multiple primary tasks are standardized to eliminate differences in units. For example, positive indicators such as timeliness and data value are normalized to the range of 0-1, while negative indicators such as resource consumption are normalized after taking their reciprocals. Then, by setting a weighted summation setting such as a 50% weight for timeliness and a 30% weight for data value, a comprehensive value score for each task is calculated, resulting in multiple primary task value scores, which form a preliminary priority ranking basis.

[0077] The multiple first benchmark tasks are sorted in descending order according to the multiple first task value scores to generate an initial priority queue, which serves as the first initial priority. It should be understood that the first initial priority here only depends on the characteristics of the task itself and does not consider the dependencies between tasks.

[0078] Based on this, this embodiment introduces a task dependency chain to correct the priority relationship. Specifically, the task dependency chain of the multiple first baseline tasks is invoked. For example, if task B needs to wait for the output of task A, the first initial priority is dynamically corrected by priority increment, and the first baseline priority is output.

[0079] If a task has a high-priority task in its dependency chain, its priority is increased according to the rules, such as its own score plus 30% of the upstream task's score.

[0080] For example, if routine task B depends on emergency task A, then the priority of B is adjusted from 0.5 to 0.5 + 0.9 × 0.3 = 0.77, and upgraded to high priority.

[0081] By repeating this process, all baseline task nodes are traversed, and the feature extraction, score calculation, and dependency correction processes are repeated to finally output the baseline task priority distribution that represents the global priority distribution map.

[0082] Step A500: Using the baseline task priority distribution as a constraint, perform container scheduling timing analysis on the baseline task node distribution and output the container resource pre-call distribution.

[0083] In one implementation, using the baseline task priority distribution as a constraint, a container scheduling timing analysis is performed on the baseline task node distribution to output the container resource pre-call distribution. Step A500 of the method includes:

[0084] Step A510: Invoke multiple first resource requirement vectors of the multiple first benchmark tasks, wherein the resource requirement vectors include container capacity requirements, computing power requirements, memory requirements, and I / O bandwidth requirements.

[0085] Step A520: Perform heterogeneous container matching based on the multiple first resource demand vectors to obtain multiple heterogeneous container codes.

[0086] Step A530: Execute container scheduling timing deployment based on coding consistency and the first baseline priority, and output the first container resource pre-call rule.

[0087] Step A540: Similarly, using the baseline task priority distribution as a constraint, perform container scheduling timing analysis on the baseline task node distribution and output the container resource pre-call distribution.

[0088] Specifically, multiple first resource requirement vectors of the multiple first benchmark tasks are invoked, and the resource requirement vectors specifically include container capability requirements, computing power requirements, memory requirements, and I / O bandwidth requirements.

[0089] The resource requirement vector quantifies the specific resource needs of a task for containers. For example, compute-intensive tasks require high-performance containers, while memory-intensive tasks require large memory containers. It also defines I / O bandwidth requirements to match data throughput. These requirement vectors are inherited from the task characteristics of resource consumption and processing time in step A410, ensuring precise matching of resource allocation with priority and load characteristics.

[0090] Based on the task resource requirement vector and the predefined heterogeneous container capability tags, such as Type-C for compute-intensive and Type-M for memory-intensive, a matching rule is established.

[0091] For example, if the task requirement is [high computing power, low memory], then a Type-C container will be matched. Each matching result generates a unique container code, identifying the binding relationship between the task and the container, and providing an atomic operation unit for subsequent scheduling.

[0092] Finally, heterogeneous containers are matched based on the multiple first resource demand vectors to obtain multiple heterogeneous container codes.

[0093] Within the same time window, the spatiotemporal consistency of container encoding is verified to avoid resource conflicts. For example, if a container is occupied by task A between 08:00 and 08:15, task B needs to be allocated to other idle containers or have its startup delayed. According to the task priority in step A440, high-priority tasks are given priority to occupy high-quality resources, such as exclusive containers for Type-P, while low-priority tasks are allocated to shared resources, such as Type-B. Finally, the first container resource pre-call rule is generated. The first container resource pre-call rule clarifies the startup time, duration, and task binding list of each execution heterogeneous container allocated to the first baseline task node when processing multiple first baseline tasks in the first baseline task node.

[0094] Similarly, with the baseline task priority distribution as a constraint, container scheduling timing analysis is performed on the baseline task node distribution to output the container resource pre-call distribution that represents the global container resource pre-call distribution.

[0095] The container resource pre-allocation distribution uses a spatiotemporal grid as a framework, marking the container type, quantity, and task load within each time slice. For example, 10 Type-C containers and 5 Type-M containers are pre-allocated in area A during UTC 08:00-10:00.

[0096] Step A600: After defining the high-load node sequence based on the high-load spatiotemporal node distribution according to the container call time window, project the high-load node sequence onto the container resource pre-call distribution to locate the container resource pre-call sequence;

[0097] Specifically, based on the container resource pre-call distribution generated in step A500, and combined with the high-load spatiotemporal node distribution in step A200, the high-load node sequence is defined by time window.

[0098] The high-load node sequence is sorted by load intensity and priority to form a core target set to be scheduled. A spatiotemporal mapping algorithm is used to match the spatiotemporal coordinates of the high-load node sequence with the locations of container resources in the pre-call distribution, such as the Type-C container group allocated in that time period, to locate the container resource pre-call sequence.

[0099] For example, if a high-load node needs to process data between 08:00 and 08:30, then 10 Type-C containers deployed during this time period are extracted from the pre-call distribution and marked as "pre-call sequence S001". This process constructs a spatiotemporal-container mapping relationship through a resource scheduling optimization algorithm, forming a pre-call strategy matrix to ensure that each high-load node has a matching container resource pool ready to serve during the predicted time period.

[0100] Step A700: After determining the load demand of the real-time carbon satellite data stream, the container resource pre-call sequence is triggered to perform time-series elastic scheduling of container resources.

[0101] When the real-time carbon satellite data stream reaches the predicted high-load spatiotemporal node, such as when data from the industrial zone begins to be transmitted at UTC 08:00A, the real-time monitoring system dynamically assesses the current load demand and compares it with the container resource supply in the pre-call sequence. If the actual load matches the prediction, such as a CPU utilization deviation of <10%, container resources are activated according to the pre-call sequence; if the load exceeds expectations, such as a sudden task causing a 30% surge in CPU demand, a dynamic elastic scheduling mechanism is triggered.

[0102] The elastic scheduling mechanism, based on a preset container pool expansion strategy, sends elastic scaling requests to the cloud container management platform, triggering a rapid instantiation process for pre-configured container images. The container orchestration engine automatically pulls preset resource configuration templates, such as a Type-E configuration with 4 core CPUs and 16GB of memory, and starts container instances in parallel across multiple cloud service provider nodes. It completes network configuration, service registration, and task queue access within a short time, achieving second-level horizontal scaling of computing resources.

[0103] Newly expanded container instances are dynamically added to the load balancing pool, and backlogged tasks on high-load nodes are preferentially distributed according to priority policies. At the same time, health checks and traffic monitoring ensure instance stability and maintain service continuity while filling the computing power gap.

[0104] This embodiment accurately predicts the distribution of high-load nodes in the carbon satellite ground system through spatiotemporal modeling, constructs a multi-dimensional resource pre-scheduling strategy and an elastic feedback closed-loop control mechanism, achieves efficient resource matching driven by dynamic load, significantly improves the utilization rate of heterogeneous container resources, and ensures the timeliness of carbon satellite monitoring data processing and the stability of the entire system link.

[0105] 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 variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for dynamic resource allocation of a carbon satellite ground system based on operational awareness, characterized in that, include: Perform load spatiotemporal correlation analysis on historical mission data of Fengyun satellites accessed from the cloud, and output a load spatiotemporal distribution cloud map; Cross-regional load conduction identification is performed on the load spatiotemporal distribution cloud map, and the high load spatiotemporal node distribution is output. The baseline task node distribution of the high-load spatiotemporal node distribution is located by spatiotemporal projection matching. After collecting task feature vectors from the baseline task node distribution, task priority analysis is performed based on the vector labels to output the baseline task priority distribution. Using the baseline task priority distribution as a constraint, perform container scheduling timing analysis on the baseline task node distribution and output the container resource pre-call distribution; After defining the high-load node sequence based on the container call time window in the high-load spatiotemporal node distribution, the high-load node sequence is projected onto the container resource pre-call distribution to locate the container resource pre-call sequence. After determining the load demand of the real-time carbon satellite data stream, the container resource pre-call sequence is triggered to perform time-series elastic scheduling of container resources. The process includes identifying cross-regional load propagation in the load spatiotemporal distribution cloud map and outputting the distribution of high-load spatiotemporal nodes, including: Based on the industrial distribution characteristics of carbon emissions, the spatiotemporal distribution cloud map of the load is divided into multiple key observation areas, and a spatial adjacency matrix between the multiple key observation areas is constructed. A temporal correlation analysis of load intensity between regions is performed on the multiple key observation areas along the spatial adjacency matrix to obtain the cross-regional load transmission path topology, wherein the transmission delay is marked on the out-degree topology connection line; A predefined load intensity threshold is used to traverse the load spatiotemporal distribution cloud map to locate the distribution of candidate high-load nodes; Based on the cross-regional load transmission path topology, the distribution of candidate high-load nodes is compensated and corrected for transmission effects to identify the additional high-load node distribution caused by cross-regional transmission. The candidate high-load node distribution and the additional high-load node distribution are spatiotemporally aligned to output the high-load spatiotemporal node distribution.

2. The method for dynamic resource allocation of a carbon satellite ground system based on operational awareness as described in claim 1, characterized in that, Perform load spatiotemporal correlation analysis on historical mission data of Fengyun satellites accessed from the cloud, and output a load spatiotemporal distribution cloud map, including: Based on the storage years, historical mission data of Fengyun satellites were decomposed to obtain multi-year carbon satellite mission data. Spatiotemporal data modeling was performed on the multi-year carbon satellite mission data to obtain multiple three-dimensional load matrices; Based on the load intensity visualization mapping rules, the load intensity mapping of the multiple three-dimensional load matrices is converted into color gradients to generate multiple historical load spatiotemporal distributions; After time-series alignment and overlay of the spatiotemporal distributions of the multiple historical loads, the spatiotemporal correlation analysis is used to uncover the periodic load transmission patterns and output the spatiotemporal distribution cloud map of the loads.

3. The method for dynamic resource allocation of a carbon satellite ground system based on operational awareness as described in claim 2, characterized in that, Spatiotemporal data modeling was performed on the multi-year carbon satellite mission data to obtain multiple three-dimensional payload matrices, including: The data from the first year's carbon satellite mission were decomposed to obtain multiple mission attribute information. A spatiotemporal grid model is constructed using the index composition of the aforementioned task attribute information as the axis. The multiple task attribute information is filled into the spatiotemporal grid model to obtain the first three-dimensional load matrix; After performing attribute decomposition on the multi-year carbon satellite mission data, the spatiotemporal grid model is reused and filled to obtain the multiple three-dimensional load matrices.

4. The method for dynamic resource allocation of a carbon satellite ground system based on operational awareness as described in claim 1, characterized in that, After collecting task feature vectors from the baseline task node distribution, task priority analysis is performed based on the vector labels to output the baseline task priority distribution, including: Extract multiple first task feature vectors from multiple first benchmark tasks in the first benchmark task node; The multiple first task features are normalized to obtain multiple first task value scores; Based on the value scores of the multiple first tasks, the multiple first benchmark tasks are serialized to obtain a first initial priority; Based on the task dependency chain of the multiple first benchmark tasks, the first initial priority is dynamically corrected by priority increment, and the first benchmark priority is output. Similarly, after collecting task feature vectors from the baseline task node distribution, task priority analysis is performed based on the vector labels to output the baseline task priority distribution.

5. The method for dynamic resource allocation of a carbon satellite ground system based on operational awareness as described in claim 4, characterized in that, The task feature vector is composed of task timeliness, data value, resource consumption, and processing time.

6. The method for dynamic resource allocation of a carbon satellite ground system based on operational awareness as described in claim 5, characterized in that, Using the baseline task priority distribution as a constraint, perform container scheduling timing analysis on the baseline task node distribution to output the container resource pre-call distribution, including: Invoke multiple first resource requirement vectors of the multiple first benchmark tasks, wherein the resource requirement vectors include container capacity requirements, computing power requirements, memory requirements and I / O bandwidth requirements; Heterogeneous container matching is performed based on the multiple first resource demand vectors to obtain multiple heterogeneous container codes; Based on coding consistency and the first baseline priority, container scheduling and deployment are performed, and the first container resource pre-call rule is output. Similarly, using the baseline task priority distribution as a constraint, container scheduling timing analysis is performed on the baseline task node distribution to output the container resource pre-call distribution.

7. The method for dynamic resource allocation of a carbon satellite ground system based on operational awareness as described in claim 3, characterized in that, The task attribute information includes task time, data stream source space, and quantified task load.

Images