A cloud resource prediction and scheduling method, system, device and medium

Abstract

This application relates to a cloud resource prediction and scheduling method, system, device, and medium. The method includes: first, acquiring real-time load data and historical load records of each node in the cloud, and inputting them into a Long Short-Term Memory (LSTM) network for processing to obtain cloud resource state fluctuation characteristics; calculating resource competition index values ​​for cross-regional nodes based on the fluctuation characteristics to determine the potential impact range of sudden peaks in cloud resource usage; using a reinforcement learning model to simulate and iteratively optimize the cloud resource allocation strategy within the specified range, outputting a global resource allocation scheme; extracting the resource allocation ratio parameters for each node's computation and storage from the scheme to generate local scheduling instructions; acquiring real-time feedback data of the local scheduling instructions, inputting it into a resource demand prediction model to determine the resource demand trend for the next cycle, and integrating these to obtain a resource adjustment scheme adapted to dynamic changes. This method can improve cloud resource utilization and enhance the system's responsiveness and operational stability in the face of load fluctuations.

Description

Technical Field

[0001] This invention belongs to the field of resource scheduling technology, and in particular relates to a cloud-based resource prediction and scheduling method, system, device and medium. Background Technology

[0002] With the widespread application of cloud computing and distributed architecture, the deployment scale of cross-regional cloud nodes is constantly expanding. The dynamic demands of various online services for computing, storage, and network resources are continuously rising. Accurate prediction and efficient scheduling of cloud resources have become crucial for ensuring business stability and improving resource utilization. Existing cloud resource scheduling methods often rely solely on real-time node load data for simple allocation, making it difficult to combine historical load information to uncover resource fluctuation patterns. They lack effective quantitative analysis of resource competition among cross-regional nodes and cannot predict the impact of sudden resource spikes. Furthermore, traditional solutions do not employ differentiated adaptation logic for the load fluctuation intensity of different nodes, resulting in a disconnect between global resource planning and local scheduling execution. This generally leads to problems such as delayed scheduling response, poor load balancing, and unreasonable resource allocation, making it difficult to meet the actual needs of large-scale, highly dynamic, cross-regional cloud services for accurate and stable resource scheduling. Summary of the Invention

[0003] Therefore, it is necessary to provide a cloud resource prediction and scheduling method, system, device, and medium that can adapt to changes in resource demand in advance, improve the overall utilization of cloud resources, and enhance the system's responsiveness to load fluctuations and operational stability in response to the above-mentioned technical problems.

[0004] Firstly, this application provides a cloud resource predictive scheduling method, including:

[0005] The system acquires real-time load data and historical load records from each node in the cloud, inputs the real-time load data and historical load records into a long short-term memory network for processing, and obtains the characteristics of cloud resource status fluctuations.

[0006] Based on the characteristics of cloud resource status fluctuations, calculate the resource competition index values ​​of cross-regional nodes, and determine the potential impact range of sudden peaks in cloud resources based on the resource competition index values.

[0007] A reinforcement learning model is used to simulate and optimize the allocation of cloud resources within the potential impact range, and output a global resource allocation scheme.

[0008] Based on the global resource allocation scheme, the allocation ratio parameters of computing resources and storage resources are extracted. If the difference between the allocation ratio parameters and the current actual allocation status exceeds a preset threshold, a local scheduling instruction is generated.

[0009] The system acquires real-time feedback data from local scheduling instructions, inputs this data into a pre-built resource demand prediction model to determine the resource demand trend for the next cycle, and integrates the global resource allocation scheme to obtain a resource adjustment scheme.

[0010] In one embodiment, resource competition index values ​​for cross-regional nodes are calculated based on cloud resource status fluctuation characteristics. The potential impact range of sudden peaks in cloud resource activity is then determined based on these resource competition index values, including:

[0011] Based on the characteristics of cloud resource status fluctuations and the resource supply of nodes across regions, the resource competition index value of nodes across regions is calculated through a preset quantification algorithm.

[0012] The resource competition index is used to quantitatively assess the tension and conflict risk in resource allocation among cross-regional nodes.

[0013] The resource competition index value is compared with the preset threshold to determine whether a sudden peak in resource demand is triggered.

[0014] If a resource surge warning is triggered, then the real-time load data of local nodes in the region corresponding to the resource competition index value is extracted.

[0015] Based on the business priority of local nodes, real-time load data is assigned fusion weights. The fused and weighted real-time load data is compared with the fluctuation characteristics of cloud resource status. Spatial correlation analysis is used to determine the potential impact range of sudden peaks in cloud resources.

[0016] In one embodiment, the resource competition index value of cross-regional nodes is calculated using the following formula:

[0017]

[0018] in, This represents the resource competition index value of cross-regional nodes. This represents the total number of cross-regional nodes participating in the calculation, determined by the actual number of nodes deployed in the cloud. Indicates the first The intensity of cloud resource status fluctuations across cross-regional nodes. Indicates the first The resource demand coefficient of each cross-regional node reflects the degree of resource demand tension corresponding to the fluctuation characteristics. , This represents the fluctuation-demand correlation coefficient, which can be dynamically calibrated based on the actual cloud scenario. Indicates the first The business priority weights of each cross-regional node are dynamically allocated based on the coreness of the business carried by the node. Indicates the first The real-time resource supply of each cross-regional node, that is, the total amount of computing, storage, and network resources that the node can currently schedule. Indicates the first Resource supply stability coefficient of cross-regional nodes This represents the cross-regional resource synergy impact coefficient.

[0019] In one embodiment, a reinforcement learning model is used to simulate and optimize the allocation of cloud resources within the potential impact range, outputting a global resource allocation scheme, including:

[0020] Obtain data on the potential impact range of cloud resource allocation; the potential impact range data includes the distribution data of resource demand and load fluctuation information of each cloud node within the range.

[0021] Based on the potential impact range data, a pre-set reinforcement learning model is used to simulate and optimize the cloud resource allocation process within the range, generating a preliminary resource allocation plan and initial optimization results.

[0022] The initial optimization results include the computation, storage, and network resource allocation parameters for each node.

[0023] The dynamic response characteristics of resource allocation for each cloud node are analyzed based on the initial optimization results, and the load balancing index is quantitatively determined based on the dynamic response characteristics.

[0024] Dynamic response characteristics include resource allocation adjustment latency and load adaptation speed; load balancing metrics are used to evaluate the degree of balance of resource load across nodes.

[0025] Determine if the load balancing metrics are below the load balancing threshold. If they are, adjust the resource allocation weights based on the business priority and load fluctuation intensity of each node to generate an updated resource allocation scheme.

[0026] Based on the resource allocation scheme, calculate the resource utilization data of each node in the cloud and the global system, and determine whether the utilization data meets the global optimization goal.

[0027] Utilization data includes computing resource utilization, storage resource utilization, and network resource utilization.

[0028] If the utilization rate data is determined not to meet the global optimization target, the resource allocation weight and resource scheduling strategy are iteratively adjusted using the utilization rate data as a feedback signal until the utilization rate data meets the global optimization target, thus obtaining a global resource allocation scheme.

[0029] In one embodiment, the allocation ratio parameters of computing resources and storage resources are extracted based on the global resource allocation scheme. If the difference between the allocation ratio parameters and the current actual allocation state exceeds a preset threshold, a local scheduling instruction is generated, including:

[0030] Based on the global resource allocation scheme, the computing and storage resource allocation ratio parameters of each cloud node and the global system are extracted in layers, and the layered ratio parameter set is obtained by associating the node business priority.

[0031] Collect the actual allocation status data of computing and storage resources of each node, compare the hierarchical ratio parameter set with the actual allocation status data node by node, calculate the ratio difference value and combine it with the load fluctuation intensity weighting to obtain the difference evaluation value.

[0032] The difference assessment value is compared with the preset difference threshold that is dynamically adapted according to business priority. If it exceeds the threshold, the network resource allocation priority is adjusted based on business priority, load fluctuation intensity and real-time network resource occupancy rate, and a preliminary local scheduling instruction is generated.

[0033] Perform a pre-verification of the load and network adaptation effect on the initial local scheduling instructions. If the verification passes, output the optimized local scheduling instructions; otherwise, adjust the network resource allocation priority until the verification passes.

[0034] In one embodiment, the difference assessment value is calculated using the following formula:

[0035]

[0036] in, Indicates the difference assessment value. This indicates the total number of cloud nodes participating in the evaluation. Indicates the first The hierarchical ratio parameter of the nth node, i.e., the nth node in the global resource allocation scheme. The ratio of computing and storage resources allocated to each node. Indicates the first The current actual allocation ratio of the nth node, i.e., the collected data of the nth node... The ratio of actual computing resources to storage resources allocated to each node. Indicates the first The business priority weight of each node Indicates the first The load fluctuation intensity weight of each node.

[0037] In one embodiment, the method further includes:

[0038] Obtain real-time feedback data of local scheduling instructions, input the real-time feedback data into a preset resource demand prediction model, and determine the cloud resource demand trend for the next cycle.

[0039] The resource demand forecasting model is a time-series forecasting model that has been pre-trained.

[0040] If the cloud resource demand trend shows that the resource load is continuously at a peak, then the global resource allocation scheme and local scheduling instructions are integrated to obtain a coordinated scheduling plan.

[0041] Resource load data is extracted from the coordination and scheduling plan, and a preset load balancing algorithm is used to determine whether the resource allocation has reached a preset threshold.

[0042] If the preset threshold is not reached, the resource allocation ratio is adjusted based on the resource load data to obtain an optimized resource scheduling strategy.

[0043] Based on the resource scheduling strategy, dynamic resource allocation instructions are generated and distributed to each cloud node for execution, thereby determining the resource adjustment plan for subsequent cycles.

[0044] Secondly, this application also provides a cloud-based resource prediction and scheduling system, the system comprising:

[0045] The load feature extraction module is used to obtain real-time load data and historical load records of each node in the cloud. The real-time load data and historical load records are input into the long short-term memory network for processing to obtain the cloud resource status fluctuation characteristics.

[0046] The resource competition analysis module is used to calculate the resource competition index values ​​of cross-regional nodes based on the fluctuation characteristics of cloud resource status, and to determine the potential impact range of sudden peaks in cloud resources based on the resource competition index values.

[0047] The global allocation optimization module is used to simulate and optimize the allocation of cloud resources within the potential impact range using a reinforcement learning model, and output a global resource allocation scheme.

[0048] The local scheduling generation module is used to extract the allocation ratio parameters of computing resources and storage resources based on the global resource allocation scheme. If the difference between the allocation ratio parameters and the current actual allocation status exceeds a preset threshold, a local scheduling instruction is generated.

[0049] The resource scheduling optimization module is used to obtain real-time feedback data of local scheduling instructions, input the real-time feedback data into the constructed resource demand prediction model to determine the resource demand trend of the next cycle, and integrate the global resource allocation scheme to obtain the resource adjustment scheme.

[0050] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0051] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned method.

[0052] The aforementioned cloud resource prediction and scheduling method, system, computer equipment, and storage medium first acquire real-time load data and historical load records of each node in the cloud. The real-time load data and historical load records are then input into a Long Short-Term Memory (LSTM) network for time-series feature extraction and processing to obtain cloud resource state fluctuation characteristics that characterize the load change patterns of nodes. Based on these cloud resource state fluctuation characteristics, combined with the i-th... The load fluctuation intensity weight of each node is used to calculate the resource competition index value of cross-regional nodes, and the potential impact range of sudden peaks in cloud resources is determined based on the distribution characteristics of the resource competition index value. A reinforcement learning model is used to simulate and iteratively optimize the cloud resource allocation strategy within the potential impact range, and output a global resource allocation scheme that adapts to global scheduling requirements. Based on the global resource allocation scheme, the allocation ratio parameters of computing resources and storage resources of each node are extracted, and the difference evaluation value between the allocation ratio parameter and the current actual allocation state of the node is calculated. If the difference evaluation value exceeds a preset threshold, a local scheduling instruction adapted to the local node's operating state is generated. Real-time feedback data after the local scheduling instruction is issued is obtained, and the real-time feedback data is input into a pre-built resource demand prediction model to determine the cloud resource demand trend in the next period. Finally, the strategy is integrated with the global resource allocation scheme to obtain a resource adjustment scheme adapted to dynamic changes. This method mines the temporal fluctuation characteristics of cloud load through Long Short-Term Memory (LSTM) networks and quantifies cross-regional resource competition relationships by combining node load fluctuation intensity weights. This enables accurate identification of the potential impact range of sudden resource peaks, improving the accuracy of resource situational awareness. Reinforcement learning is used to simulate and optimize global resource allocation, while local scheduling corrections are performed based on allocation ratio differences. This balances the rationality of global scheduling with the flexibility of local scheduling, effectively solving the problem of disconnect between global planning and local execution. It can adapt to changes in resource demand in advance, improving the overall utilization of cloud resources while enhancing the system's responsiveness and operational stability in the face of load fluctuations. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 A flowchart of a cloud resource prediction and scheduling method provided in an embodiment of the present invention;

[0055] Figure 2 This is a structural block diagram of a cloud resource prediction and scheduling system provided in an embodiment of the present invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0057] In one embodiment, such as Figure 1 As shown, this application provides a cloud resource predictive scheduling method, which may include the following steps:

[0058] Step S101: Obtain real-time load data and historical load records of each node in the cloud, input the real-time load data and historical load records into the Long Short-Term Memory network for processing, and obtain the cloud resource status fluctuation characteristics.

[0059] Specifically, the system first collects real-time load data (including computing resource utilization, data processing volume, etc.) and historical load records (covering load change data within a certain period) from each node in the cloud. These two types of data are then fed into a Long Short-Term Memory (LSTM) network for processing. This network analyzes the fluctuation amplitude and frequency of real-time load changes, as well as the historical load change patterns, extracts key features, and finally outputs the cloud resource status fluctuation characteristics (including load fluctuation amplitude and change trend, etc.).

[0060] Step S102: Calculate the resource competition index value of cross-regional nodes based on the fluctuation characteristics of cloud resource status, and determine the potential impact range of sudden peaks in cloud resources based on the resource competition index value.

[0061] Specifically, a pre-defined quantification algorithm transforms fluctuation characteristics into calculable values. Combined with the business priorities of each node, a resource competition index is calculated, directly reflecting the resource demand intensity of each node. Simultaneously, based on the distribution of the resource competition index, the potential impact range of sudden peaks in cloud resource usage is determined, identifying which nodes belong to areas of intense resource competition.

[0062] Step S103: Use a reinforcement learning model to simulate and optimize the allocation of cloud resources within the potential impact range, and output a global resource allocation scheme.

[0063] Based on defined resource competition indicators and scope of influence, parameters such as resource supply capacity, business priority, and load fluctuation intensity of each node are input into the reinforcement learning model. Through multiple rounds of iterative simulation, the resource allocation ratio is optimized, and finally a global resource allocation scheme is output.

[0064] Step S104: Extract the allocation ratio parameters of computing resources and storage resources based on the global resource allocation scheme. If the difference between the allocation ratio parameters and the current actual allocation status exceeds a preset threshold, generate a local scheduling instruction.

[0065] The computational and storage resource allocation ratios for each node are extracted based on the global resource allocation scheme, and the difference between the current allocation state and the optimal allocation state is calculated. If this difference exceeds a preset threshold, it indicates a deviation between the current local resource allocation and the global scheme. In this case, a targeted local scheduling instruction is generated. This instruction will adjust the local resource allocation based on node load fluctuations and business priorities to ensure that the resource configuration of local nodes remains consistent with the global scheme.

[0066] Step S105: Obtain real-time feedback data of local scheduling instructions, input the real-time feedback data into the constructed resource demand prediction model to determine the resource demand trend of the next cycle, and integrate the global resource allocation scheme to obtain the resource adjustment scheme.

[0067] First, execution feedback data of local scheduling instructions is collected (including load changes of each node, instruction execution effects, etc.). This feedback data is then input into a pre-built resource demand prediction model. This model combines historical load data and current feedback information to analyze the resource demand trend for the next cycle, while also integrating the global resource allocation scheme to generate a resource adjustment plan adapted to subsequent operation.

[0068] The aforementioned cloud resource prediction and scheduling method first acquires real-time load data and historical load records of each node in the cloud, inputs them into a long short-term memory network for temporal feature extraction and processing, and obtains cloud resource state fluctuation characteristics. Based on the cloud resource state fluctuation characteristics, and combined with the node load fluctuation intensity weight, the resource competition index value of cross-regional nodes is calculated, and the potential impact range of sudden peaks in cloud resources is determined according to the distribution characteristics of the index value. A reinforcement learning model is used to simulate and iteratively optimize the cloud resource allocation strategy within the range, and outputs a global resource allocation scheme. The resource allocation ratio parameters for each node are extracted from the scheme, and the difference evaluation value between the parameter and the current actual allocation state of the node is calculated. If it exceeds a preset threshold, a local scheduling instruction is generated. The real-time feedback data of the local scheduling instruction is acquired, input into a pre-built resource demand prediction model to determine the resource demand trend for the next period, and integrated with the global resource allocation scheme to obtain a resource adjustment scheme adapted to dynamic changes. This method mines the temporal fluctuation characteristics of cloud load through Long Short-Term Memory (LSTM) networks, quantifies cross-regional resource competition relationships by combining node load fluctuation intensity weights, accurately identifies the potential impact range of sudden peaks, and improves the accuracy of resource situational awareness. It uses reinforcement learning to optimize global allocation and combines differential assessment to perform local scheduling corrections, taking into account both global rationality and local flexibility, and solving the problem of disconnect between global planning and local execution. It adapts to changes in resource demand in advance, improves cloud resource utilization, and enhances the system's responsiveness and operational stability in the face of load fluctuations.

[0069] In one embodiment, calculating the resource competition index value of cross-regional nodes based on the fluctuation characteristics of cloud resource status, and determining the potential impact range of sudden peaks in cloud resources based on the resource competition index value, may include the following steps:

[0070] Step S201: Based on the fluctuation characteristics of cloud resource status and the resource supply of each node across regions, the resource competition index value of the cross-regional nodes is calculated by a preset quantification algorithm.

[0071] Preferably, the resource competition index value is used to quantitatively assess the tension and conflict risk of resource allocation among cross-regional nodes.

[0072] Step S202: Compare the resource competition index value with the preset threshold to determine whether a sudden peak in resource demand is triggered.

[0073] Step S203: If it is determined that a resource surge warning has been triggered, then extract the real-time load data of local nodes in the region corresponding to the resource competition index value.

[0074] Step S204: Assign fusion weights to real-time load data based on the business priority of local nodes. The fused and weighted real-time load data and the fluctuation characteristics of cloud resource status are used to determine the potential impact range of sudden peaks in cloud resources through spatial correlation analysis.

[0075] Specifically, based on the fluctuation characteristics of cloud resource status and the real-time resource supply of each node, a preset quantification algorithm is used to calculate the resource competition index value of cross-regional nodes. This index value is used to accurately quantify the tension and conflict risk of resource allocation between cross-regional nodes. The calculated resource competition index value is compared with a preset warning threshold to determine whether a cloud resource sudden peak warning is triggered. If a warning is triggered, the real-time load data of local nodes in the region corresponding to the warning is extracted. According to the business priority of each node, the real-time load data is assigned a fusion weight. Then, the fused and weighted real-time load data is fused with the cloud resource status fluctuation characteristics. Through spatial correlation analysis, the potential impact range of the cloud resource sudden peak is accurately determined.

[0076] This embodiment calculates resource competition indicators by combining the characteristics of cloud resource status fluctuations with the supply of node resources, achieving precise quantification of the tension and conflict risk of cross-regional node resource allocation, and improving the timeliness and accuracy of early warning of sudden resource peaks. By merging weights with business priority allocation and spatial correlation analysis to determine the potential impact range, it avoids the problem of ambiguous impact range definition, providing accurate scope basis and data support for subsequent resource scheduling optimization. It effectively solves the shortcomings of inaccurate prediction of sudden peaks and ambiguous impact range definition in traditional resource scheduling, ensuring the rationality and stability of cloud resource allocation, while improving the targeting and efficiency of resource scheduling, and ensuring the stable operation of the cloud system.

[0077] In one embodiment, the resource competition index value of cross-regional nodes can be calculated using the following formula:

[0078]

[0079] in, This represents the resource competition index value of cross-regional nodes. This represents the total number of cross-regional nodes participating in the calculation, determined by the actual number of nodes deployed in the cloud. Indicates the first The intensity of cloud resource status fluctuations across cross-regional nodes. Indicates the first The resource demand coefficient of each cross-regional node reflects the degree of resource demand tension corresponding to the fluctuation characteristics. , This represents the fluctuation-demand correlation coefficient, which can be dynamically calibrated based on the actual cloud scenario. Indicates the first The business priority weights of each cross-regional node are dynamically allocated based on the coreness of the business carried by the node. Indicates the first The real-time resource supply of each cross-regional node, that is, the total amount of computing, storage, and network resources that the node can currently schedule. Indicates the first Resource supply stability coefficient of cross-regional nodes This represents the cross-regional resource synergy impact coefficient.

[0080] This embodiment integrates multi-dimensional data such as node load fluctuation parameters, business priority weights, resource supply and stability coefficients to accurately quantify the resource competition relationship between cross-regional nodes, avoiding the one-sidedness of traditional single-parameter calculations. It can clearly reflect the tension of resource supply and demand at each node, providing scientific data support for predicting sudden resource peaks and defining the scope of impact. At the same time, it enables the quantification and traceability of resource competition status, ensuring the accuracy and reliability of resource competition index values. This provides a precise basis for subsequent resource allocation optimization and scheduling instruction generation, effectively improving the rationality and pertinence of cloud resource scheduling, ensuring that resource allocation matches the actual operating needs of nodes, reducing resource waste, enhancing the stability of cloud system operation and resource utilization efficiency, and solving the problems of vague traditional resource competition assessment and lack of data support for scheduling.

[0081] In one embodiment, employing a reinforcement learning model to simulate and optimize the allocation of cloud resources within the potential impact range, and outputting a global resource allocation scheme, may include the following steps:

[0082] Step S301: Obtain data on the potential impact range of cloud resource allocation; the potential impact range data includes the resource demand distribution data and load fluctuation information of each cloud node within the range.

[0083] Step S302: Based on the potential impact range data, a preset reinforcement learning model is used to simulate and optimize the cloud resource allocation process within the range, generating a preliminary resource allocation scheme and initial optimization results.

[0084] Preferably, the initial optimization results include the computation, storage, and network resource allocation parameters for each node.

[0085] The core input is the data on the potential impact range of sudden peaks in cloud resources. This potential impact range data covers the resource demand distribution data and load fluctuation information of all cloud nodes within the range. The resource demand distribution data is used to characterize the scale and distribution characteristics of each node's demand for computing, storage and network resources, while the load fluctuation information is used to reflect the magnitude and trend of load changes of each node during operation.

[0086] The aforementioned potential impact range data is imported into a pre-built reinforcement learning model. Node resource scheduling constraints serve as the optimization boundary, including upper limits on node hardware resources and minimum resource guarantee requirements for business operation. Based on the input resource demand and load fluctuation information, the reinforcement learning model conducts multiple iterative simulations of the cloud resource allocation process within this potential impact range. In each iteration, the allocation ratio and method of computing, storage, and network resources for each node are adaptively adjusted, gradually selecting an allocation method suitable for the current resource situation. After multiple rounds of iterative optimization, the model outputs a preliminary resource allocation scheme adapted to the resource operation status of this potential impact range, and simultaneously generates corresponding initial optimization results.

[0087] Step S304: Analyze the dynamic response characteristics of resource allocation for each cloud node based on the initial optimization results, and quantify and determine the load balancing index based on the dynamic response characteristics.

[0088] Dynamic response characteristics include resource allocation adjustment latency and load adaptation speed; load balancing metrics are used to evaluate the degree of balance of resource load across nodes.

[0089] Step S305: Determine whether the load balancing index is lower than the load balancing threshold. If it is lower, adjust the resource allocation weights according to the business priority and load fluctuation intensity of each node to generate an updated resource allocation scheme.

[0090] Step S306: Calculate the resource utilization data of each node in the cloud and the global resource allocation scheme, and determine whether the utilization data meets the global optimization target.

[0091] Utilization data includes computing resource utilization, storage resource utilization, and network resource utilization.

[0092] Step S307: If it is determined that the utilization rate data has not reached the global optimization target, the resource allocation weight and resource scheduling strategy are iteratively adjusted using the utilization rate data as a feedback signal until the utilization rate data meets the global optimization target, and a global resource allocation scheme is obtained.

[0093] Specifically, firstly, data on the potential impact range of cloud resource allocation is obtained. This data includes resource demand distribution data and load fluctuation information for each cloud node within the corresponding range. Based on this data, a preset reinforcement learning model is used to simulate and optimize the cloud resource allocation process within this range, generating a preliminary resource allocation scheme and obtaining initial optimization results. These initial optimization results include computational, storage, and network resource allocation parameters for each node. For the initial optimization results, the dynamic response characteristics of resource allocation for each cloud node are analyzed. These dynamic response characteristics include resource allocation adjustment latency and load adaptation speed. Based on these dynamic response characteristics, a load balancing index is quantified to evaluate the load balancing degree of each node. The load balancing index is compared with a preset load balancing threshold. If the load balancing index is lower than the threshold, the resource allocation weight is adjusted according to the business priority and load fluctuation intensity of each node, generating an updated resource allocation scheme. Based on the updated resource allocation scheme, the resource utilization data of each node in the cloud and the global resource utilization data are calculated. The resource utilization data includes computing resource utilization, storage resource utilization, and network resource utilization. It is then determined whether the utilization data meets the preset global optimization target. If the utilization data does not meet the global optimization target, the utilization data is used as a feedback signal to iteratively adjust the resource allocation weight and resource scheduling strategy until the utilization data meets the global optimization target, and finally the global resource allocation scheme is obtained.

[0094] This embodiment achieves preliminary optimization of resource allocation by combining potential impact range data with reinforcement learning models, providing a reasonable initial allocation basis for cloud resource scheduling. By quantifying load balancing indicators based on dynamic response characteristics and adjusting allocation weights based on node business priorities and load fluctuation intensity, the balance and node adaptability of resource allocation can be improved. This can effectively improve the overall utilization efficiency of computing, storage, and network resources, while enhancing the dynamic response capability of cloud resource scheduling to load changes, ensuring the global rationality of resource allocation and system operational stability.

[0095] In one embodiment, the allocation ratio parameter of computing resources and storage resources is extracted based on the global resource allocation scheme. If the difference between the allocation ratio parameter and the current actual allocation state exceeds a preset threshold, a local scheduling instruction is generated, which may include the following steps:

[0096] Step S401: Based on the global resource allocation scheme, extract the computing and storage resource allocation ratio parameters of each cloud node and the global system, and associate them with the node business priority to obtain the hierarchical ratio parameter set.

[0097] A hierarchical extraction approach is adopted to extract computing and storage resource allocation ratios for each cloud node and the overall system from both the global planning and single-node execution levels. Global-level parameters constrain overall resource allocation, while node-level parameters characterize the resource allocation baseline for individual nodes. The extracted node allocation ratios are then correlated with the corresponding node's business priority, and integrated using a pre-defined weighted formula, which can be expressed as: In the formula For the first The hierarchical ratio parameter of each node. The first one extracted from the global resource allocation scheme The basic allocation ratio parameter for each node. For the first The business priority weights of each node are then extracted and associated with priorities as described above, ultimately forming a hierarchical ratio parameter set that includes global constraints and node-specific configurations.

[0098] Step S402: Collect the actual allocation status data of computing and storage resources of each node, compare the hierarchical ratio parameter set with the actual allocation status data node by node, calculate the ratio difference value and combine it with the load fluctuation intensity weighting to obtain the difference evaluation value.

[0099] Step S403: Compare the difference assessment value with the preset difference threshold that is dynamically adapted according to service priority. If the difference exceeds the threshold, adjust the network resource allocation priority based on service priority, load fluctuation intensity and real-time network resource occupancy rate, and generate preliminary local scheduling instructions.

[0100] Step S404: Perform a pre-verification of the load and network adaptation effect on the preliminary local scheduling instruction. If the verification passes, output the optimized local scheduling instruction; otherwise, adjust the network resource allocation priority until the verification passes.

[0101] For the generated preliminary local scheduling instructions, pre-verification is performed from two dimensions: node load adaptation effect and network resource adaptation effect. By simulating the instruction execution process, it is determined whether the node load is within a reasonable operating range and whether the network resource allocation can meet the business data transmission and interaction requirements. If both verification indicators meet the preset standards, the verification is deemed successful, and the preliminary local scheduling instruction is directly identified as the optimized local scheduling instruction and output. If the verification fails, it indicates that the current network resource allocation priority setting is unreasonable and needs to be readjusted. Based on the adjusted network resource allocation priority, the local scheduling instruction is regenerated, and the load and network adaptation effect pre-verification is performed again. This process is iterated until the verification conditions are met, and finally, the optimized local scheduling instruction that can be directly issued and executed is output.

[0102] Specifically, based on the global resource allocation scheme, the allocation ratio parameters of computing and storage resources for each cloud node and the global level are extracted hierarchically. Simultaneously, the business priority tags of each node are associated to form a hierarchical ratio parameter set, clarifying the resource allocation benchmark for nodes with different business priorities. The actual allocation status data of current computing and storage resources for each cloud node are collected. The allocation ratio parameters corresponding to each node in the hierarchical ratio parameter set are compared node by node with the current actual allocation status data of that node to calculate the resource allocation ratio difference value for each node. This difference value is then weighted by the load fluctuation intensity weight to obtain the resource allocation difference assessment value for each node. Finally, the difference assessment value for each node is compared with a preset difference threshold. The differentiation threshold is dynamically adapted based on the node's service priority; the higher the service priority, the stricter the threshold setting. If the difference assessment value of a node exceeds its corresponding preset differentiation threshold, the network resource allocation priority is dynamically adjusted based on the node's service priority, load fluctuation intensity, and real-time network resource occupancy rate, generating a targeted preliminary local scheduling instruction. The generated preliminary local scheduling instruction undergoes load and network adaptation effect pre-verification, simulating the node's load changes and network resource adaptation after instruction execution. If the pre-verification result meets the preset adaptation requirements, the optimized local scheduling instruction is output. If the pre-verification fails, the network resource allocation priority is continuously fine-tuned until the pre-verification passes, ensuring that the local scheduling instruction can adapt to the node's actual operating state.

[0103] This embodiment extracts resource allocation ratio parameters hierarchically and quantifies resource allocation differences by associating them with business priorities. This enables precise identification of deviations between resource allocation at each node and the global plan, improving the accuracy and relevance of difference assessment. By dynamically adapting differentiated thresholds for business priorities, it ensures that the generation of local scheduling instructions aligns with node business needs, avoiding resource waste or business interruptions caused by indiscriminate scheduling. The pre-verification process can proactively avoid potential issues such as load imbalance and network adaptation anomalies after instruction execution, ensuring the feasibility of local scheduling instructions. It achieves an organic combination of global resource planning and local node adaptation, effectively solving the problem of disconnect between global resource allocation and local node operation, improving the rationality and accuracy of cloud resource allocation, and enhancing the system's adaptability to load fluctuations, ensuring the stable and efficient operation of the cloud system.

[0104] In one embodiment, the difference assessment value can be calculated using the following formula:

[0105]

[0106] in, Indicates the difference assessment value. This indicates the total number of cloud nodes participating in the evaluation. Indicates the first The hierarchical ratio parameter of the nth node, i.e., the nth node in the global resource allocation scheme. The ratio of computing and storage resources allocated to each node. Indicates the first The current actual allocation ratio of the nth node, i.e., the collected data of the nth node... The ratio of actual computing resources to storage resources allocated to each node. Indicates the first The business priority weight of each node Indicates the first The load fluctuation intensity weight of each node.

[0107] The difference assessment value calculation formula in this embodiment combines the global planning resource allocation ratio, the actual resource allocation status of nodes, business priority, and load fluctuation intensity to achieve multi-dimensional quantification of resource allocation deviation. It can objectively and accurately reflect the degree of deviation between the resource allocation of each cloud node and the global plan. By weighted fusion of multiple influencing factors, the reliability and pertinence of the difference assessment results are improved, providing a unified and standardized judgment basis for the generation of subsequent local scheduling instructions. This helps to improve the accuracy of cloud resource scheduling and ensure that the global resource allocation plan is compatible with the actual operating status of nodes.

[0108] In one embodiment, the method may further include the following steps:

[0109] Step S501: Obtain real-time feedback data of local scheduling instructions, input the real-time feedback data into a preset resource demand prediction model, and determine the cloud resource demand trend for the next period.

[0110] Preferably, the resource demand prediction model is a pre-trained time-series prediction model.

[0111] First, real-time feedback data is collected from each cloud node during the execution of local scheduling instructions. This real-time feedback data includes the actual utilization rate of each node's computing and storage resources, instruction execution latency, load variation magnitude, and load fluctuation intensity. This real-time feedback data is then completely input into a pre-trained resource demand prediction model (time-series prediction model). The model, combining the input feedback data, analyzes and infers the time-series characteristics of load fluctuations and resource demand change patterns of each node, accurately predicting the resource demand changes of each node and the entire system in the next cycle. Ultimately, it determines the cloud resource demand trend for the next cycle, clarifying the overall trend of resource load and the increase or decrease in resource demand for each node in the next cycle.

[0112] Step S502: If the cloud resource demand trend shows that the resource load is continuously at a peak, then the global resource allocation scheme and local scheduling instructions are integrated to obtain a coordinated scheduling plan.

[0113] Specifically, the trend of cloud resource demand is first analyzed. If it is determined that the resource load of each node in the cloud will continue to be at a high level in the next cycle, the global resource allocation plan is used as the core basis for overall scheduling. At the same time, the node adaptation rules corresponding to the local scheduling instructions that have been issued and executed are combined to organically integrate the overall resource planning at the global level with the node scheduling constraints at the local level, eliminate the execution deviation between global scheduling and local scheduling, and finally form a coordinated scheduling plan that takes into account both the global resource optimization goal and the actual operating status of local nodes.

[0114] Step S503: Extract resource load data from the coordination and scheduling plan, and determine whether the resource allocation has reached the preset threshold through a preset load balancing algorithm.

[0115] In step S504, if the preset threshold is not reached, the resource allocation ratio is adjusted according to the resource load data to obtain the optimized resource scheduling strategy.

[0116] Step S505: Generate dynamic resource allocation instructions based on the resource scheduling strategy and distribute them to each cloud node for execution to determine the resource adjustment plan for subsequent periods.

[0117] Specifically, real-time feedback data after the issuance and execution of local scheduling instructions is obtained. This real-time feedback data is input into a pre-trained and constructed time-series prediction model (i.e., a resource demand prediction model). Through the model's analysis and inference of the time-series characteristics of the feedback data, the cloud resource demand trend for the next cycle is determined. If the cloud resource demand trend indicates that the resource load will continue to be at its peak, the generated global resource allocation scheme and the currently executed local scheduling instructions are integrated to sort out the core requirements of global planning and local adaptation, forming a coordinated scheduling plan that takes into account both global rationality and local adaptability. Resource load data of each cloud node is extracted from the coordinated scheduling plan and input into a preset load balancing algorithm. Through algorithm calculation and analysis, it is determined whether the current resource allocation status has reached a preset threshold. If the result is that the preset threshold has not been reached, the resource allocation ratio of each node is adjusted based on the extracted resource load data, and the resource allocation logic is optimized to obtain an optimized resource scheduling strategy. Based on this optimized resource scheduling strategy, dynamic resource allocation instructions are generated and distributed to each cloud node for execution. At the same time, combined with the current resource demand trend and instruction execution expectations, the resource adjustment scheme for subsequent cycles is determined to ensure the continuity and dynamic adaptability of resource scheduling.

[0118] This embodiment achieves closed-loop feedback of resource scheduling effectiveness by collecting real-time feedback data from local scheduling commands. Combined with a time-series prediction model, it can accurately predict resource demand trends for the next cycle, proactively address sustained peak resource loads, and avoid scheduling lag issues. By integrating the global resource allocation scheme with local scheduling commands, it balances the overall integrity of global resource planning with the targeted adaptation of local nodes, solving the problem of disconnect between global and local scheduling. Relying on load balancing algorithms to verify resource allocation effectiveness and dynamically adjust allocation ratios effectively improves the balance of resource allocation, avoiding resource waste and load imbalance. It can dynamically adapt to changes in resource demand, improve cloud resource utilization and scheduling accuracy, and ensure the stable and efficient operation of the cloud system under sustained peak load conditions.

[0119] In one embodiment, such as Figure 2 As shown, this application also provides a cloud-based resource prediction and scheduling system, which may include:

[0120] The load feature extraction module 601 is used to obtain real-time load data and historical load records of each node in the cloud, input the real-time load data and historical load records into the long short-term memory network for processing, and obtain the cloud resource status fluctuation characteristics.

[0121] The resource competition analysis module 602 is used to calculate the resource competition index value of cross-regional nodes based on the fluctuation characteristics of cloud resource status, and to determine the potential impact range of sudden peaks in cloud resources based on the resource competition index value.

[0122] The global allocation optimization module 603 is used to simulate and optimize the allocation of cloud resources within the potential impact range using a reinforcement learning model, and output a global resource allocation scheme.

[0123] The local scheduling generation module 604 is used to extract the allocation ratio parameters of computing resources and storage resources based on the global resource allocation scheme. If the difference between the allocation ratio parameters and the current actual allocation status exceeds a preset threshold, a local scheduling instruction is generated.

[0124] The resource scheduling optimization module 605 is used to obtain real-time feedback data of local scheduling instructions, input the real-time feedback data into the constructed resource demand prediction model to determine the resource demand trend of the next cycle, and integrate the global resource allocation scheme to obtain the resource adjustment scheme.

[0125] The aforementioned cloud-based resource prediction and scheduling system comprises a load feature extraction module, a resource contention analysis module, a global optimization module, and a local scheduling module that work together to construct a complete resource scheduling system. Specifically, the load feature extraction module collects real-time and historical operational data from cloud nodes, inputs it into a long short-term memory network (LSTM), and processes it using algorithms to obtain the fluctuation characteristics of cloud resource states, providing a foundation for subsequent competition index calculations. The resource contention analysis module, based on these fluctuation characteristics and combined with node business priorities, calculates resource contention indicators using a preset algorithm to clarify the degree of resource supply and demand tension. The global optimization module employs a reinforcement learning model to perform multiple rounds of simulation optimization on resource allocation, outputting a suitable resource allocation scheme. The local scheduling module, based on actual operational states, generates targeted scheduling instructions to ensure that resource allocation aligns with actual needs, achieving scientific, rational, and efficient resource allocation.

[0126] This embodiment effectively solves the problems of "emphasizing the overall situation while neglecting the local situation" and "emphasizing allocation while neglecting feedback" in traditional resource scheduling, achieving precise matching between resource allocation and the actual operating status of nodes. Through multi-module linkage, it not only ensures efficient resource utilization and avoids resource waste, but also improves the targeting and stability of resource scheduling, ensuring the efficient operation of the cloud system. At the same time, through accurate demand prediction and dynamic adjustment, it effectively responds to load fluctuations, improves the scientific nature and practicality of resource scheduling, provides solid support for the stable operation of the cloud system, and achieves a dual improvement in resource utilization efficiency and operational stability.

[0127] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0128] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the cloud resource prediction and scheduling method as described above.

[0129] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0130] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0131] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A cloud resource prediction and scheduling method, characterized in that, The method includes: The real-time load data and historical load records of each node in the cloud are obtained, and the real-time load data and historical load records are input into a long short-term memory network for processing to obtain the cloud resource status fluctuation characteristics. Based on the fluctuation characteristics of cloud resource status, calculate the resource competition index value of cross-regional nodes, and determine the potential impact range of sudden peaks in cloud resources based on the resource competition index value; A reinforcement learning model is used to simulate and optimize the allocation of cloud resources within the potential impact range, and a global resource allocation scheme is output. Based on the global resource allocation scheme, the allocation ratio parameters of computing resources and storage resources are extracted. If the difference between the allocation ratio parameters and the current actual allocation status exceeds a preset threshold, a local scheduling instruction is generated. The real-time feedback data of the local scheduling instruction is obtained, and the real-time feedback data is input into the constructed resource demand prediction model to determine the resource demand trend in the next cycle. The global resource allocation scheme is then integrated to obtain the resource adjustment scheme.

2. The method according to claim 1, characterized in that, The calculation of resource competition index values ​​for cross-regional nodes based on the fluctuation characteristics of cloud resource status, and the determination of the potential impact range of sudden peaks in cloud resource usage based on the resource competition index values, include: Based on the cloud resource status fluctuation characteristics and the resource supply of each node across regions, the resource competition index value of the cross-regional nodes is calculated through a preset quantification algorithm. The resource competition index value is used to quantitatively assess the tension and conflict risk in resource allocation among cross-regional nodes; The resource competition index value is compared with a preset threshold to determine whether a sudden peak in resource demand is triggered. If the resource surge warning is triggered, then extract the real-time load data of local nodes in the region corresponding to the resource competition index value. Based on the business priority of the local nodes, the real-time load data is assigned a fusion weight. The fused and weighted real-time load data is then compared with the cloud resource status fluctuation characteristics. Spatial correlation analysis is used to determine the potential impact range of sudden peaks in cloud resources.

3. The method according to claim 2, characterized in that, The resource competition index value of the cross-regional nodes is calculated using the following formula: in, This represents the resource competition index value of cross-regional nodes. This represents the total number of cross-regional nodes participating in the calculation, determined by the actual number of nodes deployed in the cloud. Indicates the first The intensity of cloud resource status fluctuations across cross-regional nodes. Indicates the first The resource demand coefficient of each cross-regional node reflects the degree of resource demand tension corresponding to the fluctuation characteristics. , This represents the fluctuation-demand correlation coefficient, which can be dynamically calibrated based on the actual cloud scenario. Indicates the first The business priority weights of each cross-regional node are dynamically allocated based on the coreness of the business carried by the node. Indicates the first The real-time resource supply of each cross-regional node, that is, the total amount of computing, storage, and network resources that the node can currently schedule. Indicates the first Resource supply stability coefficient of cross-regional nodes This represents the cross-regional resource synergy impact coefficient.

4. The method according to claim 1, characterized in that, The step of using a reinforcement learning model to simulate and optimize cloud resource allocation within the potential impact range, and outputting a global resource allocation scheme, includes: Obtain data on the potential impact range of cloud resource allocation; the potential impact range data includes resource demand distribution data and load fluctuation information of each cloud node within the range; Based on the potential impact range data, a preset reinforcement learning model is used to simulate and optimize the cloud resource allocation process within the range, generating a preliminary resource allocation scheme and initial optimization results; The initial optimization results include the computation, storage, and network resource allocation parameters for each node; The dynamic response characteristics of resource allocation for each cloud node are analyzed based on the initial optimization results, and load balancing indicators are quantitatively determined based on these dynamic response characteristics. The dynamic response characteristics include resource allocation adjustment latency and load adaptation speed; the load balancing metrics are used to evaluate the degree of balance of resource load across nodes. Determine whether the load balancing index is lower than the load balancing threshold. If it is, adjust the resource allocation weights according to the business priority and load fluctuation intensity of each node to generate an updated resource allocation scheme. Based on the resource allocation scheme, calculate the resource utilization data of each node in the cloud and the global resource utilization data, and determine whether the utilization data meets the global optimization target. The utilization data includes computing resource utilization, storage resource utilization, and network resource utilization; If it is determined that the utilization rate data has not reached the global optimization target, then the resource allocation weight and resource scheduling strategy are iteratively adjusted using the utilization rate data as a feedback signal until the utilization rate data meets the global optimization target, thus obtaining a global resource allocation scheme.

5. The method according to claim 1, characterized in that, The step involves extracting the allocation ratio parameters of computing and storage resources based on the global resource allocation scheme. If the difference between the allocation ratio parameters and the current actual allocation state exceeds a preset threshold, a local scheduling instruction is generated, including: Based on the global resource allocation scheme, the computing and storage resource allocation ratio parameters of each cloud node and the global system are extracted in layers, and the layered ratio parameter set is obtained by associating the node business priority. Collect the actual allocation status data of computing and storage resources of each node, compare the hierarchical ratio parameter set with the actual allocation status data node by node, calculate the ratio difference value and combine it with the load fluctuation intensity weighting to obtain the difference evaluation value; The difference assessment value is compared with the preset difference threshold that is dynamically adapted according to the service priority. If it exceeds the threshold, the network resource allocation priority is adjusted based on the service priority, load fluctuation intensity and real-time network resource occupancy rate, and a preliminary local scheduling instruction is generated. The initial local scheduling instruction is pre-verified for load and network adaptation effects. If the verification passes, the optimized local scheduling instruction is output; otherwise, the network resource allocation priority is adjusted until the verification passes.

6. The method according to claim 1, characterized in that, The difference assessment value is calculated using the following formula: in, Indicates the difference assessment value. This indicates the total number of cloud nodes participating in the evaluation. Indicates the first The hierarchical ratio parameter of the nth node, i.e., the nth node in the global resource allocation scheme. The ratio of computing and storage resources allocated to each node. Indicates the first The current actual allocation ratio of the nth node, i.e., the collected data of the nth node... The ratio of actual computing resources to storage resources allocated to each node. Indicates the first The business priority weight of each node Indicates the first The load fluctuation intensity weight of each node.

7. The method according to claim 1, characterized in that, The method further includes: Obtain real-time feedback data of the local scheduling instruction, input the real-time feedback data into a preset resource demand prediction model, and determine the cloud resource demand trend for the next period. The resource demand prediction model is a pre-trained time-series prediction model; If the cloud resource demand trend shows that the resource load is continuously at a peak, then the global resource allocation scheme and the local scheduling instructions are integrated to obtain a coordinated scheduling plan; Resource load data is extracted from the coordination and scheduling plan, and a preset load balancing algorithm is used to determine whether the resource allocation has reached a preset threshold. If the preset threshold is not reached, the resource allocation ratio is adjusted according to the resource load data to obtain an optimized resource scheduling strategy. Based on the resource scheduling strategy, dynamic resource allocation instructions are generated and distributed to each cloud node for execution, thereby determining the resource adjustment plan for subsequent periods.

8. A cloud-based resource prediction and scheduling system, characterized in that, The system includes: The load feature extraction module is used to obtain real-time load data and historical load records of each node in the cloud, and input the real-time load data and historical load records into the long short-term memory network for processing to obtain the cloud resource status fluctuation characteristics. The resource competition analysis module is used to calculate the resource competition index value of cross-regional nodes based on the fluctuation characteristics of cloud resource status, and determine the potential impact range of sudden peaks in cloud resources based on the resource competition index value. The global allocation optimization module is used to simulate and optimize the allocation of cloud resources within the potential impact range using a reinforcement learning model, and output a global resource allocation scheme. The local scheduling generation module is used to extract the allocation ratio parameters of computing resources and storage resources based on the global resource allocation scheme. If the difference between the allocation ratio parameters and the current actual allocation status exceeds a preset threshold, a local scheduling instruction is generated. The resource scheduling optimization module is used to obtain real-time feedback data of the local scheduling instructions, input the real-time feedback data into the constructed resource demand prediction model to determine the resource demand trend of the next cycle, and integrate the global resource allocation scheme to obtain the resource adjustment scheme.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images