Resource scheduling method, device, equipment, medium and computer program product

Abstract

The invention provides a resource scheduling method, device and equipment, a medium and a computer program product. The method comprises the following steps: converting received service demand information into resource demand parameters; performing correlation evaluation on the resource data information and the resource demand parameter of the scheduling node based on the resource data prediction model to obtain resources required by the service; the resource data prediction model is obtained by training an LSTM model based on historical resource data; the parameter of the resource data prediction model is an optimal parameter combination obtained by performing population iteration optimization on a particle swarm based on a particle swarm optimization algorithm by taking the loss value of the LSTM as a fitness value and the parameter of the LSTM as the particle swarm in a model training stage; and determining a resource scheduling scheme based on the resources required by the service and the resource scheduling model. According to the method, through unified node resource data management, historical resource data are trained by applying the particle swarm optimization algorithm and the LSTM model, and efficient scheduling and accurate scheduling of resource demands are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of resource scheduling, and in particular to a resource scheduling method, device, equipment, medium and computer program product. Background Art

[0002] Existing resource scheduling methods usually use historical tasks and their corresponding demand resources to construct associated data, and train prediction models through the constructed resource demand associated data to predict the resources to be scheduled corresponding to the task demand. However, existing resource scheduling methods do not fully utilize cluster resources, resulting in low resource scheduling efficiency; because demand and resources are not directly related, existing resource scheduling methods lack certain flexibility, and the prediction model based on demand resource association training does not accurately predict resources, resulting in low resource scheduling accuracy. Summary of the invention

[0003] The present invention provides a resource scheduling method, device, equipment, medium and computer program product, which are used to solve the defects of low resource scheduling efficiency and low accuracy in the existing resource scheduling technology, and improve the resource scheduling efficiency and accuracy.

[0004] The present invention provides a resource scheduling method, comprising the following steps.

[0005] Convert the received business demand information into resource demand parameters; Based on the resource data prediction model, the resource data information of the scheduling node and the resource demand parameters are evaluated for correlation to obtain the resources required for the business; the resource data prediction model is obtained by training the LSTM model based on historical resource data; the parameters of the resource data prediction model are obtained by taking the loss value of the LSTM model as the fitness value and the parameters of the LSTM model as the particle swarm in the model training stage, and performing population iterative optimization on the particle swarm based on the particle swarm optimization algorithm to obtain the optimal parameter combination; A resource scheduling solution is determined based on the resources required for the business and a resource scheduling model; the resource scheduling model is trained based on historical demand data and the historical resource data.

[0006] According to a resource scheduling method provided by the present invention, converting the received service demand information into resource demand parameters comprises: Determine multiple sub-demand sets based on the received business demand information; Inputting the target sub-requirement in each of the sub-requirement sets into the large language model based on the thinking chain prompting algorithm, and obtaining the reasoning chain and resource combination of each of the target sub-requirements output by the large language model; Based on the reasoning chain and resource combination of each of the target sub-requirements, the resource requirement parameters corresponding to the business requirement information are determined.

[0007] According to a resource scheduling method provided by the present invention, the resource scheduling method further includes: Splitting the historical demand data to obtain multiple historical sub-demands; Based on the historical sub-resources corresponding to each of the historical sub-demands, a historical reasoning chain is constructed; the thinking chain prompt algorithm is obtained based on the training of the historical reasoning chain; The large language model is obtained based on historical sub-demands, historical sub-resources and historical reasoning chain training with corresponding relationships.

[0008] According to a resource scheduling method provided by the present invention, the scheduling node includes a main regulation point, a backup regulation point and an auxiliary regulation point; the source scheduling method also includes: Associating the main regulation point, the backup regulation point and the auxiliary regulation point based on the node distance and the service demand to obtain node association information; Determine the master node to be scheduled based on the resource requirement parameter; When the resource utilization rate of the master node to be scheduled is higher than a preset threshold and the master node to be scheduled does not meet the service demand information, determining a backup node to be scheduled and a secondary node to be scheduled based on the node association information; Based on the main node to be scheduled, the backup node to be scheduled and the auxiliary node to be scheduled, resource data information of the scheduling node is determined.

[0009] According to a resource scheduling method provided by the present invention, the resource scheduling method further includes: The learning rate, the number of hidden layer nodes and the regularization parameter of the LSTM model are used as optimization objects of the particle swarm optimization algorithm; Update the particle swarm based on the particle position, particle velocity and global optimal position, apply the parameters corresponding to the updated particle swarm to the LSTM model, and obtain an updated fitness value; When the updated fitness value meets the population iterative optimization requirement, the LSTM model is trained based on the parameters corresponding to the updated particle swarm to obtain the resource data prediction model; In the case that the updated fitness value does not meet the population iterative optimization requirement, return to the step of updating the particle swarm and the fitness value until the updated fitness value meets the population iterative optimization requirement.

[0010] According to a resource scheduling method provided by the present invention, the resource scheduling method further includes: Determine historical resource information that meets the historical demand data; Mapping and matching the historical resource information with historical node resources to obtain multiple resource scheduling algorithms; Based on the feedback information corresponding to each of the resource scheduling algorithms, a resource scheduling solution that meets the historical demand data is determined.

[0011] The present invention also provides a resource scheduling device, comprising the following modules: A business requirement information conversion module, used to convert the received business requirement information into resource requirement parameters; A module for determining resources required for a business, which is used to evaluate the correlation between the resource data information of the scheduling node and the resource demand parameters based on a resource data prediction model to obtain the resources required for the business; the resource data prediction model is obtained by training the LSTM model based on historical resource data; the parameters of the resource data prediction model are obtained by taking the loss value of the LSTM model as the fitness value and the parameters of the LSTM model as the particle swarm in the model training stage, and performing population iterative optimization on the particle swarm based on the particle swarm optimization algorithm to obtain the optimal parameter combination; The resource scheduling scheme determination module is used to determine the resource scheduling scheme based on the resources required by the business and the resource scheduling model; the resource scheduling model is trained based on the historical demand data and the historical resource data.

[0012] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any of the resource scheduling methods described above when executing the computer program.

[0013] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the resource scheduling method described in any one of the above is implemented.

[0014] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the resource scheduling methods described above.

[0015] The resource scheduling method, device, equipment, medium and computer program product provided by the present invention achieve accurate prediction of resource demand through unified node resource data management, and then apply particle swarm optimization algorithm and LSTM model to train historical resource data, thereby improving resource scheduling efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 This is one of the flow charts of the resource scheduling method provided by the present invention.

[0018] Figure 2 This is the second flow chart of the resource scheduling method provided by the present invention.

[0019] Figure 3 It is a structural diagram of the resource scheduling device provided by the present invention.

[0020] Figure 4 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] Combine the following Figure 1-Figure 4 The resource scheduling method, apparatus, device, medium and computer program product of the present invention are described.

[0023] Figure 1 It is one of the flow charts of the resource scheduling method provided by the present invention, such as Figure 1 As shown, the method includes the following: Step 100: Convert the received business demand information into resource demand parameters; After receiving the multimodal business demand information input by the user, the business demand information is converted into text information, and semantic thinking chain prompt information is constructed based on the business scenario. The converted text information is further split into multiple business needs through the thinking chain prompt method CoT, and the resource requirements of each business requirement are parsed one by one through the pre-trained large language model, and finally the structured resource requirement parameters are obtained.

[0024] Step 200: Based on the resource data prediction model, the resource data information of the scheduling node and the resource demand parameters are evaluated for correlation to obtain the resources required for the business; the resource data prediction model is obtained by training the LSTM model based on historical resource data; the parameters of the resource data prediction model are obtained by performing population iterative optimization on the particle swarm based on the particle swarm optimization algorithm in the model training stage, using the loss value of the LSTM model as the fitness value and the parameters of the LSTM model as the particle swarm to obtain the optimal parameter combination; When evaluating user needs, a flexible resource scheduling node structure is proposed through the pre-built cluster resource data management platform, and various resource data of nodes (such as AI resources, computing resources, and network resources, etc.) are collected. Then, various types of historical resource data are clustered, and then these historical resource data are trained through the particle swarm optimization_long short-term memory algorithm (Particle Swarm Optimization_Long Short-Term Memory, PSO_LSTM) to generate a resource data prediction model. The resource data prediction model can predict the resource status of each scheduling node based on the resource data information of each current scheduling node, and evaluate the correlation between the predicted resource status of each current scheduling node and user needs to obtain the specific resources corresponding to the business needs, that is, the resources required for the business.

[0025] Step 300: Determine a resource scheduling solution based on the resources required by the business and a resource scheduling model; the resource scheduling model is trained based on historical demand data and the historical resource data.

[0026] Specifically, after training the resource data prediction model, a demand-resource mapping template is constructed based on structured resource demand parameters to determine the resource category, resource specification, and resource quantity that meet the business scenario requirements. The resource information is then mapped and matched with the resources managed by each node in the cluster resource data management platform, and finally a highly adaptable resource scheduling algorithm is determined. The feasibility of the resource scheduling algorithm is verified, and a certain number of resource scheduling solutions are generated and recommended to users for selection.

[0027] For example, when the user's business demand information only includes high-definition video, but does not specify specific resource requirements, or the user is not clear about what resources are needed to support high-definition video, in this case, according to the set demand and resource mapping template, 500M bandwidth resources can be matched to support high-definition video. Or if the user's business demand information includes the replay function, according to the demand and resource mapping template, multi-day cloud storage resources can be matched to support the realization of the demand.

[0028] This embodiment achieves accurate prediction of resource demand through unified node resource data management, and then applies particle swarm optimization algorithm and LSTM model to train historical resource data, thereby improving resource scheduling efficiency and accuracy.

[0029] Figure 2 This is the second flow chart of the resource scheduling method provided by the present invention. Figure 2 As shown, the method may also include: Step 110: determining a plurality of sub-demand sets based on the received business demand information; Step 120: input the target sub-demand in each of the sub-demand sets into the large language model based on the thought chain prompting algorithm, and obtain the reasoning chain and resource combination of each of the target sub-demand output by the large language model; Step 130: Determine resource requirement parameters corresponding to the business requirement information based on the reasoning chain and resource combination of each target sub-requirement.

[0030] In order to further illustrate the conversion process from business demand information to resource demand parameters, a thinking chain demand analysis system is proposed. The thinking chain demand analysis system can efficiently analyze user needs and convert user needs into structured resource demand parameters.

[0031] The Thinking Chain Demand Parsing System breaks down complex problems into multiple sub-problems through a step-by-step parsing method, and then parses them one by one. CoT can realize the conversion from input to thinking chain to output. A prompt containing CoT consists of three key parts: instructions, logical basis, and examples. The instructions define the problem and indicate the output format; the logical basis part includes problem-solving strategies, reasoning steps, and related knowledge; and the example part determines the input and output format of the large language model. Each example contains a question, reasoning process, and answer.

[0032] The Thinking Chain Demand Analysis System splits user requirements according to different business resource requirements to obtain multiple sub-requirements, and then classifies the multiple sub-requirements, and classifies the sub-requirements of the same business resource requirements into one category. Add the clustered sub-requirements to the requirement set, and then sample the business sub-requirements, and select a representative sub-requirement from each requirement set. Input the sub-requirement into the large language model to generate the reasoning chain and the corresponding resources. When all sub-requirements are completed, a sub-requirement sampling list containing multiple business requirements + reasoning chains + resources is obtained. Finally, all sub-requirement samples are spliced ​​to generate new business requirements, build new prompts, and then input the prompts into the large language model to obtain the resource requirement parameters corresponding to the business requirement information.

[0033] This embodiment constructs semantic thinking chain prompt information based on the content of actual business application scenarios, uses CoT to assist the large language model to efficiently parse business needs, and completes the conversion of user natural language expressions into structured resource demand parameters.

[0034] In one embodiment, the resource scheduling method provided in the embodiment of the present application may further include: Step 10: Split the historical demand data to obtain multiple historical sub-demands; Step 20: construct a historical reasoning chain based on the historical sub-resources corresponding to each of the historical sub-demands; the thinking chain prompt algorithm is obtained based on the training of the historical reasoning chain; Step 30: The large language model is obtained based on the historical sub-demands, historical sub-resources and historical reasoning chain training that have corresponding relationships.

[0035] Specifically, from the above content, it can be seen that the thinking chain prompt algorithm can cooperate with the large language model, and the sub-demands obtained by splitting the demand data are input into the large language model, and the large language model will output the reasoning chain corresponding to the sub-demand and the corresponding demand resources. It can be seen that the large language model is obtained through the training of the association relationship between sub-demand + reasoning chain + demand resources. The reasoning chain is generated by the thinking chain prompt algorithm obtained through the training of the association relationship between user needs and corresponding demand resources. The thinking chain prompt algorithm consists of three key parts: instructions, logical basis and examples. By continuously learning examples and logical basis, the thinking chain prompt algorithm can determine the reasoning chain between the two based on the input user needs and corresponding demand resources, and assist in the training of the large language model.

[0036] This embodiment assists the training of the large language model through a thought chain prompt algorithm.

[0037] In one embodiment, the resource scheduling method provided in the embodiment of the present application may further include: Step 40: Associating the main regulation point, the backup regulation point and the auxiliary regulation point based on the node distance and the service demand to obtain node association information; Step 50: Determine the master node to be scheduled based on the resource demand parameter; Step 60: When the resource utilization rate of the master node to be scheduled is higher than a preset threshold and the master node to be scheduled does not meet the service demand information, determine a backup node to be scheduled and an auxiliary node to be scheduled based on the node association information; Step 70: Determine resource data information of the scheduling node based on the to-be-scheduled master node, the to-be-scheduled backup node, and the to-be-scheduled auxiliary node.

[0038] Specifically, the present application also proposes a unified cluster data management platform, through which a 1+N+X scheduling node structure is constructed, where 1 is the main scheduling node; N is the backup scheduling node; and X is the auxiliary scheduling node. According to resource utilization, each scheduling node can be divided into three types: high resource utilization, medium resource utilization, and low resource utilization. The scheduling nodes are mainly associated according to the distance and business needs (for example, in the scenario of east-west computing, data in the eastern region is stored in the western region), which facilitates real-time resource scheduling and ensures the quality of resource support.

[0039] After converting business demand information into resource demand parameters through the Thinking Chain Demand Analysis System, the scheduling node can directly schedule related resources based on the resource demand parameters. When the resource utilization of the main node is high but does not meet business needs, the business requirements such as distance and performance can be calculated based on the resource utilization of the corresponding associated scheduling nodes to schedule the resources of other backup scheduling nodes and auxiliary scheduling nodes.

[0040] Moreover, the resources managed by the scheduling nodes of the 1+N+X structure correspond one-to-one to the actual resource conditions of the area it manages, so the resources of each scheduling node are not consistent. For example, a certain AI capability is only available in the auxiliary scheduling node, but not in other scheduling nodes. In this case, the resources of the auxiliary scheduling node must be scheduled to meet the user's business needs. Therefore, each scheduling node is closely related.

[0041] This embodiment solves the problems of data dispersion and low utilization efficiency and realizes data integration by building a unified cluster resource data management platform.

[0042] In one embodiment, the resource scheduling method provided in the embodiment of the present application may further include: Step 210: taking the learning rate, the number of hidden layer nodes and the regularization parameter of the LSTM model as optimization objects of the particle swarm optimization algorithm; Step 220: updating the particle swarm based on the particle position, particle velocity and global optimal position, applying the parameters corresponding to the updated particle swarm to the LSTM model to obtain an updated fitness value; Step 230: When the updated fitness value meets the population iterative optimization requirement, the LSTM model is trained based on the parameters corresponding to the updated particle swarm to obtain the resource data prediction model; Step 240: If the updated fitness value does not meet the population iterative optimization requirement, return to the step of updating the particle swarm and the fitness value until the updated fitness value meets the population iterative optimization requirement.

[0043] Specifically, the training process of the above resource data prediction model is as follows.

[0044] The collected historical resource data are clustered and divided into proportions to obtain training historical data, verification historical data, and test historical data. The training of the resource data prediction model is based on the LSTM model, and the PSO algorithm can achieve the optimal parameter configuration of the LSTM model. The PSO algorithm can search for the optimal combination of weights and bias parameters in the LSTM model. By continuously adjusting the parameters, the LSTM model can better fit the training data and have better generalization capabilities.

[0045] When training the resource data prediction model, the learning rate, number of hidden layer nodes, and regularization parameters of the LSTM model are used as optimization objects, the PSO algorithm is initialized, and the error between the model prediction result and the actual result is calculated as the fitness value. To automatically adjust the LSTM parameters through the PSO algorithm, the particle swarm must first be initialized, and a group of particles are randomly generated. Each particle can represent a set of LSTM model parameter values. According to the current particle position and speed and the global optimal position, the particle position and speed are updated, and then the parameters corresponding to the updated particle position are applied to the LSTM model, and the fitness value is further calculated and updated, and the particle speed and global optimal position are continued to be updated. After continuous iterative optimization (the fitness value reaches the expected value or the number of iterations reaches the preset value), the optimal hyperparameters are output. The LSTM model is then trained with the optimal hyperparameters. The data prediction model is trained with training data and verification data, and then predicted with the test set to obtain the final resource data prediction results.

[0046] Through this training, the resource data prediction model can combine historical resource data and resource status to achieve accurate prediction of resource status and provide reasonable and efficient resources for business needs.

[0047] This embodiment automatically adjusts LSTM parameters through the PSO algorithm, trains various types of historical resource data, generates a resource data prediction model, and then predicts the resource status of each scheduling node based on the resource data information of the current scheduling nodes, thereby improving the accuracy of resource scheduling.

[0048] In one embodiment, the resource scheduling method provided in the embodiment of the present application may further include: Step 400: Determine historical resource information that meets the historical demand data; Step 500: Map and match the historical resource information with historical node resources to obtain multiple resource scheduling algorithms; Step 600: Determine a resource scheduling solution that meets the historical demand data based on feedback information corresponding to each resource scheduling algorithm.

[0049] Specifically, the common scheduling algorithms currently include the First Come First Served (FCFS) scheduling algorithm, the Short Job First (SJF) scheduling algorithm, and the High Response Ratio Priority (HRRN) scheduling algorithm. These scheduling algorithms are not suitable for computing network service resource scheduling scenarios, and do not fully consider the impact of factors such as user level and service level on the scheduling order in actual service scenarios. For actual computing network service scenarios, this application proposes an efficient resource scheduling algorithm. First, a time range is set, and a demand set is generated for the resource requirements of different users within the time range. The first demand sorting is performed according to the time node, and then the demand set is judged, mainly to calculate the resource demand in the demand set, and evaluate whether the resource status of the current scheduling node is sufficient to support the full realization of the business requirements in the set, so as to facilitate the use of different resource scheduling methods to schedule resources. If the current node resources are sufficient to support the full realization of the business in the set, it is necessary to assign levels to the business requirements and calculate the complexity of the set requirements in the demand set, as shown in Formula 1.

[0050] ; (1) In formula 1, user levels can be divided into general users, important users and core users, and different levels are assigned corresponding weights. For example, core users are assigned weights of 1.5, important users are assigned weights of 1.0, and general users are assigned weights of 0.8. If there are core users in a set, the set is directly upgraded to a high-complexity set; if there are multiple core users in a set, they are sorted according to the importance of their needs. Service levels can be divided into single service, hybrid service and integrated service, and different levels of services are assigned corresponding weights, for example, single service is 1.1, hybrid service is 1.0, and integrated service is 0.9.

[0051] After the calculation is completed, the complexity of the set requirements is judged. If the complexity of the set requirements is greater than the set threshold, it is defined as a high-complexity set. The scheduling order is to process the core user requirements first and sort them according to the importance of the requirements; then process the single business and mixed business requirements and sort them according to the importance of the requirements; and process the remaining requirements one by one according to the time nodes. If the complexity of the set requirements is less than the threshold, it is defined as a low-complexity set. For a low-complexity set, it is necessary to calculate the importance of each requirement in the set and sort them, and then process the business requirements one by one in order. The importance of the requirements can refer to four factors: time node (the earlier the value, the larger the value), urgency (the more urgent the value, the larger the value, which can be obtained based on user demand analysis), business level and user level. For example, the time node weight can be set to 0.3, and the urgency weight can be set to 0.1, as shown in Formula 2.

[0052] ; (2) If the current node resources are not sufficient to support the realization of all the services in the set, it is necessary to filter out the subset of business requirements that can be supported according to the resource situation, calculate the importance of each requirement in the business requirement subset, and then sort the services in the subset according to the importance of the requirements, and process the requirements one by one. For business requirements outside the subset, the resources of the corresponding alternative nodes and auxiliary nodes can be scheduled according to standards such as distance, performance, and service support. When it is perceived again that the resources of the main scheduling node can support the realization of services outside the subset, the PSO_LSTM algorithm can be used to train historical business data to generate a prediction model, and select the appropriate time period to migrate business support resources to the managed resources of the main scheduling node according to the prediction model.

[0053] The user feedback optimization part is mainly based on the user feedback after the actual resource scheduling. The resource scheduling plan can be appropriately adjusted according to the user feedback, and the user feedback can be passed to the user demand receiving part to re-analyze the user demand and make resource predictions. Then, the scheduling algorithm can be used to re-schedule the (computing network) resources that meet the user's latest needs.

[0054] This embodiment reasonably controls the issuance of resource scheduling instructions by combining the two indicators of demand complexity and demand importance to achieve optimal resource allocation, and continuously adjusts and optimizes resource scheduling through user feedback.

[0055] The resource scheduling device provided by the present invention is described below. The resource scheduling device described below and the resource scheduling method described above can be referenced to each other.

[0056] Please refer to Figure 3 The present invention also provides a resource scheduling device, comprising: The business requirement information conversion module 301 is used to convert the received business requirement information into resource requirement parameters; The business required resource determination module 302 is used to evaluate the correlation between the resource data information of the scheduling node and the resource demand parameters based on the resource data prediction model to obtain the resources required for the business; the resource data prediction model is obtained by training the LSTM model based on historical resource data; the parameters of the resource data prediction model are the optimal parameter combination obtained by performing population iterative optimization on the particle swarm based on the particle swarm optimization algorithm in the model training stage, using the loss value of the LSTM model as the fitness value and the parameters of the LSTM model as the particle swarm; The resource scheduling scheme determination module 303 is used to determine a resource scheduling scheme based on the resources required by the business and a resource scheduling model; the resource scheduling model is trained based on historical demand data and the historical resource data.

[0057] Optionally, the business requirement information conversion module includes: A sub-requirement set determining unit, configured to determine a plurality of sub-requirement sets based on the received business requirement information; An inference chain and resource combination output unit, used to input the target sub-requirement in each sub-requirement set into the large language model based on a thinking chain prompting algorithm, and obtain the inference chain and resource combination of each target sub-requirement output by the large language model; The resource requirement parameter determination unit is used to determine the resource requirement parameters corresponding to the business requirement information based on the reasoning chain and resource combination of each target sub-requirement.

[0058] Optionally, the resource scheduling device further includes: A historical demand data splitting module is used to split the historical demand data to obtain multiple historical sub-demands; A historical reasoning chain construction module is used to construct a historical reasoning chain based on the historical sub-resources corresponding to each of the historical sub-demands; the thinking chain prompt algorithm is obtained based on the training of the historical reasoning chain; The large language model training module is used to obtain the large language model through training based on historical sub-demands, historical sub-resources and historical reasoning chains that have corresponding relationships.

[0059] Optionally, the scheduling node includes a main regulation point, a backup regulation point and an auxiliary regulation point; and the resource scheduling device further includes: A node association module, used to associate the main regulation point, the backup regulation point and the auxiliary regulation point based on the node distance and the service demand to obtain node association information; A module for determining a master node to be scheduled, used to determine a master node to be scheduled based on the resource demand parameters; A standby and auxiliary node determination module, configured to determine a standby node and an auxiliary node to be scheduled based on the node association information when the resource utilization rate of the main node to be scheduled is higher than a preset threshold and the main node to be scheduled does not meet the service demand information; The resource data information determination module is used to determine the resource data information of the scheduling node based on the main node to be scheduled, the backup node to be scheduled and the auxiliary node to be scheduled.

[0060] Optionally, the resource scheduling device further includes: An optimization object determination module, used to use the learning rate, the number of hidden layer nodes and the regularization parameter of the LSTM model as the optimization object of the particle swarm optimization algorithm; A particle swarm updating module is used to update the particle swarm based on the particle position, particle velocity and global optimal position, and apply the parameters corresponding to the updated particle swarm to the LSTM model to obtain an updated fitness value; A model training module module is used to train the LSTM model based on the parameters corresponding to the updated particle swarm to obtain the resource data prediction model when the updated fitness value meets the population iterative optimization requirements; The loop module is used to return to the step of updating the particle swarm and the fitness value when the updated fitness value does not meet the population iterative optimization requirement, until the updated fitness value meets the population iterative optimization requirement.

[0061] Optionally, the resource scheduling device further includes: A historical resource information determination module, used to determine historical resource information that meets the historical demand data; A mapping and matching module, used to map and match the historical resource information with historical node resources to obtain multiple resource scheduling algorithms; The resource scheduling scheme determination module is used to determine a resource scheduling scheme that meets the historical demand data based on feedback information corresponding to each of the resource scheduling algorithms.

[0062] Figure 4 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 4 As shown, the electronic device may include: a processor 410, a communication interface 420, a memory 430 and a communication bus 440, wherein the processor 410, the communication interface 420 and the memory 430 communicate with each other through the communication bus 440. The processor 410 may call the logic instructions in the memory 430 to execute the resource scheduling method, which includes: converting the received business demand information into resource demand parameters; performing correlation evaluation on the resource data information of the scheduling node and the resource demand parameters based on the resource data prediction model to obtain the resources required for the business; the resource data prediction model is obtained by training the LSTM model based on historical resource data; the parameters of the resource data prediction model are the optimal parameter combination obtained by performing population iteration optimization on the particle swarm based on the particle swarm optimization algorithm in the model training stage, using the loss value of the LSTM model as the fitness value and the parameters of the LSTM model as the particle swarm; determining the resource scheduling scheme based on the resources required for the business and the resource scheduling model; the resource scheduling model is obtained by training based on the historical demand data and the historical resource data.

[0063] In addition, the logic instructions in the above-mentioned memory 430 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0064] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the resource scheduling method provided by the above methods, which method includes: converting the received business demand information into resource demand parameters; performing a correlation evaluation on the resource data information of the scheduling node and the resource demand parameters based on a resource data prediction model to obtain the resources required for the business; the resource data prediction model is obtained by training an LSTM model based on historical resource data; the parameters of the resource data prediction model are the optimal parameter combination obtained by performing population iterative optimization on the particle swarm based on a particle swarm optimization algorithm during the model training phase, using the loss value of the LSTM model as the fitness value and the parameters of the LSTM model as the particle swarm; based on the resources required for the business and the resource scheduling model, determining a resource scheduling plan; the resource scheduling model is obtained by training based on historical demand data and the historical resource data.

[0065] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the resource scheduling method provided by the above-mentioned methods, the method comprising: converting the received business demand information into resource demand parameters; performing a correlation evaluation on the resource data information of the scheduling node and the resource demand parameters based on a resource data prediction model to obtain the resources required for the business; the resource data prediction model is obtained by training an LSTM model based on historical resource data; the parameters of the resource data prediction model are obtained by performing population iterative optimization of the particle swarm based on a particle swarm optimization algorithm during the model training phase, using the loss value of the LSTM model as the fitness value and the parameters of the LSTM model as the particle swarm; determining a resource scheduling plan based on the resources required for the business and the resource scheduling model; the resource scheduling model is obtained by training based on historical demand data and the historical resource data.

[0066] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0067] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0068] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A resource scheduling method, characterized in that: include: Convert the received business demand information into resource demand parameters; Based on the resource data prediction model, the resource data information of the scheduling node and the resource demand parameters are evaluated for correlation to obtain the resources required for the business; the resource data prediction model is obtained by training the LSTM model based on historical resource data; the parameters of the resource data prediction model are obtained by taking the loss value of the LSTM model as the fitness value and the parameters of the LSTM model as the particle swarm in the model training stage, and performing population iterative optimization on the particle swarm based on the particle swarm optimization algorithm to obtain the optimal parameter combination; A resource scheduling solution is determined based on the resources required for the business and a resource scheduling model; the resource scheduling model is trained based on historical demand data and the historical resource data.

2. The resource scheduling method according to claim 1, characterized in that: The converting the received business requirement information into resource requirement parameters comprises: Determine multiple sub-demand sets based on the received business demand information; Inputting the target sub-requirement in each of the sub-requirement sets into the large language model based on the thinking chain prompting algorithm, and obtaining the reasoning chain and resource combination of each of the target sub-requirements output by the large language model; Based on the reasoning chain and resource combination of each of the target sub-requirements, the resource requirement parameters corresponding to the business requirement information are determined.

3. The resource scheduling method according to claim 2, characterized in that: The resource scheduling method further includes: Splitting the historical demand data to obtain multiple historical sub-demands; Based on the historical sub-resources corresponding to each of the historical sub-demands, a historical reasoning chain is constructed; the thinking chain prompt algorithm is obtained based on the training of the historical reasoning chain; The large language model is obtained based on historical sub-demands, historical sub-resources and historical reasoning chain training with corresponding relationships.

4. The resource scheduling method according to claim 1, characterized in that: The scheduling nodes include a main regulation point, a backup regulation point and an auxiliary regulation point; The source scheduling method further includes: Associating the main regulation point, the backup regulation point and the auxiliary regulation point based on the node distance and the service demand to obtain node association information; Determine the master node to be scheduled based on the resource requirement parameter; When the resource utilization rate of the master node to be scheduled is higher than a preset threshold and the master node to be scheduled does not meet the service demand information, determining a backup node to be scheduled and a secondary node to be scheduled based on the node association information; Based on the main node to be scheduled, the backup node to be scheduled and the auxiliary node to be scheduled, resource data information of the scheduling node is determined.

5. The resource scheduling method according to claim 1, characterized in that: The resource scheduling method further includes: The learning rate, the number of hidden layer nodes and the regularization parameter of the LSTM model are used as optimization objects of the particle swarm optimization algorithm; Update the particle swarm based on the particle position, particle velocity and global optimal position, apply the parameters corresponding to the updated particle swarm to the LSTM model, and obtain an updated fitness value; When the updated fitness value meets the population iterative optimization requirement, the LSTM model is trained based on the parameters corresponding to the updated particle swarm to obtain the resource data prediction model; In the case that the updated fitness value does not meet the population iterative optimization requirement, return to the step of updating the particle swarm and the fitness value until the updated fitness value meets the population iterative optimization requirement.

6. The resource scheduling method according to claim 1, characterized in that: The resource scheduling method further includes: Determine historical resource information that meets the historical demand data; Mapping and matching the historical resource information with historical node resources to obtain multiple resource scheduling algorithms; Based on the feedback information corresponding to each of the resource scheduling algorithms, a resource scheduling solution that meets the historical demand data is determined.

7. A resource scheduling device, characterized in that: include: A business requirement information conversion module, used to convert the received business requirement information into resource requirement parameters; A module for determining resources required for a business, which is used to evaluate the correlation between the resource data information of the scheduling node and the resource demand parameters based on a resource data prediction model to obtain the resources required for the business; the resource data prediction model is obtained by training the LSTM model based on historical resource data; the parameters of the resource data prediction model are obtained by taking the loss value of the LSTM model as the fitness value and the parameters of the LSTM model as the particle swarm in the model training stage, and performing population iterative optimization on the particle swarm based on the particle swarm optimization algorithm to obtain the optimal parameter combination; A resource scheduling scheme determination module, used to determine a resource scheduling scheme based on the resources required by the business and the resource scheduling model; The resource scheduling model is trained based on historical demand data and historical resource data.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the resource scheduling method according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the resource scheduling method according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the resource scheduling method according to any one of claims 1 to 6 is implemented.

Images