Cloud phone resource dynamic scheduling method, device, electronic device and storage medium

Abstract

The present invention relates to a method, device, electronic device and storage medium for dynamic scheduling of cloud phone resources, the method comprising: receiving a first request from a user terminal, the first request being divided into an instance allocation request and an instance exit request. When the first request is an instance allocation request, a safety threshold for the number of idle cloud phone instances at each level is formulated according to historical allocation data in response to the instance allocation request to determine whether the idle instance pool in the real-time multi-cloud resource cluster has a cloud phone instance that can be safely allocated. When there is a cloud phone instance that can be safely allocated in the idle instance pool, the optimal instance allocation result is obtained according to the status of multiple cloud resources through a deep reinforcement learning algorithm, the corresponding cloud phone instance resources are allocated to the user terminal, and a neural network model is called to extract the user terminal's usage habit characteristics for the cloud phone instance resources. Based on the usage habit characteristics, the cloud phone instance resources are recycled to the idle instance pool or the cloud phone instance resources are suspended in the working instance pool.

Description

Technical Field

[0001] The present invention relates to the technical field of resource allocation, and in particular to a method, device, electronic device and storage medium for dynamic scheduling of cloud phone resources. Background Art

[0002] In the resource allocation of cloud phones, the cloud resource center has limited cloud resources. Users can request to use cloud phone resources for a specific time in chronological order, and the cloud resource service center needs to decide whether to accept the user's application or temporarily put the user's application on hold. For users who are allocated cloud phone instance resources, most of them will not use cloud phones 24 hours a day, so there are a lot of idle resource periods, resulting in resource waste. If there are still a lot of idle resources on the physical server, dynamic resource scheduling generally does not have a significant impact on the business running in the virtual machine. However, if these cloud phone instances are over-allocated and the resources of the physical server are close to saturation, then the cloud phone instances will compete with each other and wait for each other, thereby increasing the delay of the business in each virtual machine. Therefore, reasonable cloud resource allocation can maximize the quality of service to users and improve user satisfaction under limited resources.

[0003] At present, most existing resource allocation methods use heuristic algorithms for resource allocation. However, the behavior patterns of users using cloud phones are relatively changeable, which makes the duration of each cloud phone call by users uncertain. In addition, the usage period has a significant time-varying characteristic and has a daily periodic pattern. For example, some services are mainly used during working hours, and some services are mainly used in spare time. The overall usage of each day is roughly the same. It is difficult for heuristic algorithms to achieve good results in such a complex scenario. In addition, the existing resource allocation method is also implemented by scheduling and optimizing the resource supply cluster of the same cloud platform, but this method is difficult to coordinate the scheduling of cloud resource instances across multiple cloud platforms, and it is inconvenient to perform optimal allocation of cloud resources on multiple cloud platforms. In addition, considering that reinforcement learning can directly learn how to make better decisions from the experience of interacting with the environment, the resource clusters of cloud phone instances of different levels provided by multiple cloud platforms are relatively complex, and the user's usage behavior habits are relatively random, making the feasibility of state transition probability modeling poor.

[0004] In summary, the existing cloud phone resource allocation methods are highly dependent on the cloud platform, resulting in low cloud resource utilization efficiency. In addition, due to the limitations of data quality on reinforcement learning, the intelligence and flexibility of resource allocation strategies are low. Summary of the invention

[0005] Based on this, it is necessary to provide a cloud phone resource dynamic scheduling method, device, electronic device and storage medium that can improve the efficiency of cloud resource utilization and the intelligence and flexibility of resource allocation strategies in response to the above technical problems.

[0006] The present invention provides a cloud phone resource dynamic scheduling method, which is applied to a dynamic resource management system of a cloud resource center, and the method comprises:

[0007] receiving a first request from a user terminal, wherein the first request is divided into an instance allocation request and an instance exit request;

[0008] When the first request is an instance allocation request, in response to the instance allocation request, a safety threshold for the number of idle cloud phone instances at each level is formulated according to historical allocation data to determine whether an idle instance pool in a real-time multi-cloud resource cluster has a cloud phone instance that can be safely allocated;

[0009] When there is a cloud phone instance that can be safely allocated in the idle instance pool of the real-time multi-cloud resource cluster, based on the instance allocation request and the status of multiple cloud resources in the idle instance pool of the current multi-cloud resource cluster, an optimal instance allocation result is obtained according to the status of the multiple cloud resources through a deep reinforcement learning algorithm;

[0010] Allocate corresponding cloud phone instance resources to the user terminal according to the optimal instance allocation result, and call a neural network model to extract the user terminal's usage habit characteristics of the cloud phone instance resources according to the instance exit request;

[0011] Based on the usage habit characteristics of the cloud phone instance resources by the user end, the cloud phone instance resources are recycled to the idle instance pool according to the allocation source, or the cloud phone instance resources are suspended in the working instance pool;

[0012] Among them, the safely allocated cloud phone instances are cloud phone instance resources in which the number of idle cloud phone instances in each level is not less than the safety threshold, the deep reinforcement learning algorithm is used to train allocation strategies corresponding to different cloud phone instance resources according to different reward targets corresponding to the multiple cloud resources, and the usage habit characteristics include the user's use time period, usage duration and resource usage of the allocated cloud phone instance resources.

[0013] In one embodiment, when the first request is an instance allocation request, in response to the instance allocation request, a safety threshold for the number of idle cloud phone instances at each level is formulated according to historical allocation data to determine whether an idle instance pool in a real-time multi-cloud resource cluster has a cloud phone instance that can be safely allocated, and the above also includes:

[0014] When the first request is an instance exit request, calling the neural network model to predict the user's usage habit characteristics for the current cloud phone instance resources;

[0015] Based on the user's usage habits of the current cloud phone instance resources, the current cloud phone instance resources are recovered to the corresponding idle instance pool according to the allocation source, or the current cloud phone instance resources are suspended in the working instance pool.

[0016] In one embodiment, when the first request is an instance allocation request, in response to the instance allocation request, a safety threshold for the number of idle cloud phone instances at each level is formulated according to historical allocation data to determine whether an idle instance pool in a real-time multi-cloud resource cluster has a cloud phone instance that can be safely allocated, including:

[0017] According to the safety threshold, it is determined whether the idle instance pool in the real-time multi-cloud resource cluster has a cloud phone instance that can be safely allocated; if so,

[0018] In response to the instance allocation request, obtain the optimal instance allocation result according to the corresponding allocation strategy; if not, then

[0019] After the first period of time, it is determined again based on the safety threshold whether there is a cloud phone instance that can be safely allocated in the idle instance pool, so as to wait for the cloud phone instance that can be safely allocated.

[0020] In one of the embodiments, when there is a cloud phone instance that can be safely allocated in the idle instance pool of the real-time multi-cloud resource cluster, based on the instance allocation request and the status of multiple cloud resources in the idle instance pool of the current multi-cloud resource cluster, the optimal instance allocation result is obtained according to the status of the multiple cloud resources through a deep reinforcement learning algorithm, including:

[0021] Obtaining a reward target corresponding to each cloud resource in the real-time multi-cloud resource cluster, and training allocation strategies corresponding to different reward targets according to each cloud resource and the reward target corresponding to each cloud resource through the deep reinforcement learning algorithm;

[0022] Calling the trained deep reinforcement learning algorithm to process the states of the various cloud resources in the idle instance pool to output optimal allocation results corresponding to the states of the various cloud resources;

[0023] Among them, the optimal allocation result is the cloud phone instance resource with the highest reward target among the multiple cloud resources.

[0024] In one embodiment, the allocating corresponding cloud phone instance resources to the user terminal according to the optimal instance allocation result, and calling a neural network model to extract the user terminal's usage habit characteristics of the cloud phone instance resources according to the instance exit request, includes:

[0025] Obtaining usage habit features of multiple cloud phone instance resources, and using the multiple cloud phone instance resources as inputs of the neural network model, using the usage habit features corresponding to each cloud phone instance resource as outputs of the neural network model, and training the neural network model;

[0026] Based on the instance exit request, the trained neural network model is called to process the allocated cloud phone instance resources to predict the usage habit characteristics corresponding to the allocated cloud phone instance resources.

[0027] In one embodiment, the recovering the cloud phone instance resources to the idle instance pool according to the allocation source based on the usage habit characteristics of the cloud phone instance resources by the user terminal, or suspending the cloud phone instance resources in the working instance pool, includes:

[0028] Determine whether the cloud phone instance resource is used again within the second time period according to the usage habit characteristics corresponding to the allocated cloud phone instance resource; if so,

[0029] Suspend the allocated cloud phone instance resources in the working instance pool; if not,

[0030] The allocated cloud phone instance resources are recovered to the original idle instance pool according to the allocation source.

[0031] In one embodiment, the method further comprises:

[0032] When the allocated cloud phone instance resources are recycled to the idle instance pool and the corresponding cloud phone instance resources are allocated to the user terminal according to the optimal instance allocation result, the corresponding cloud resource status in the real-time multi-cloud resource cluster is dynamically updated.

[0033] The present invention also provides a cloud phone resource dynamic scheduling device, which is applied to a dynamic resource management system of a cloud resource center, and the device comprises:

[0034] A request receiving module, used to receive a first request from a user terminal, wherein the first request is divided into an instance allocation request and an instance exit request;

[0035] A request response module, for, when the first request is an instance allocation request, formulating safety thresholds for the number of idle cloud phone instances at various levels according to historical allocation data in response to the instance allocation request, so as to determine whether an idle instance pool in a real-time multi-cloud resource cluster has a cloud phone instance that can be safely allocated;

[0036] A resource allocation module, configured to obtain an optimal instance allocation result according to the status of multiple cloud resources by using a deep reinforcement learning algorithm based on the instance allocation request and the status of multiple cloud resources in the idle instance pool of the current multi-cloud resource cluster when there is a cloud phone instance that can be safely allocated in the idle instance pool of the real-time multi-cloud resource cluster;

[0037] A user behavior prediction module is used to allocate corresponding cloud phone instance resources to the user terminal according to the optimal instance allocation result, and call a neural network model to extract the user terminal's usage habit characteristics of the cloud phone instance resources according to the instance exit request;

[0038] A resource processing module, configured to recycle the cloud phone instance resources to the idle instance pool according to the allocation source, or suspend the cloud phone instance resources in the working instance pool based on the usage habit characteristics of the cloud phone instance resources by the user end;

[0039] Among them, the safely allocated cloud phone instances are cloud phone instance resources in which the number of idle cloud phone instances in each level is not less than the safety threshold, the deep reinforcement learning algorithm is used to train allocation strategies corresponding to different cloud phone instance resources according to different reward targets corresponding to the multiple cloud resources, and the usage habit characteristics include the user's use time period, usage duration and resource usage of the allocated cloud phone instance resources.

[0040] The present invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the cloud phone resource dynamic scheduling method as described in any one of the above.

[0041] The present invention also provides a computer storage medium storing a computer program, which, when executed by a processor, implements any of the cloud phone resource dynamic scheduling methods described above.

[0042] The above-mentioned cloud phone resource dynamic scheduling method, device, electronic device and storage medium use a neural network model to extract user characteristics and predict the user's usage habits such as the time period, usage duration and resource usage of the cloud phone. Moreover, the use of the neural network model can more accurately predict the user's behavior habits, thereby effectively planning cloud phone resources, and can also make full use of the resources of different cloud platforms to improve resource utilization efficiency, while reducing dependence on a single cloud platform. The resource allocation strategy of the cluster is optimized by using deep reinforcement learning. By analyzing the historical scheduling data in the multi-cloud resource cluster and the various time resource usage patterns of the cloud phone instance, the security threshold of the multi-cloud resource cluster can be predicted according to the real-time situation, and the cloud resource instance can be dynamically adjusted, which improves the intelligence and flexibility of the resource allocation strategy and ensures the safe operation of the system. Therefore, the method can make full use of the resources of different cloud platforms, realize the intelligent management and optimization of cloud phone instance resources, maximize resource utilization efficiency, reduce costs, and provide a good user experience while meeting user needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] 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.

[0044] Figure 1 This is a schematic diagram of a method for dynamically scheduling cloud phone resources provided by the present invention;

[0045] Figure 2 A schematic diagram of the cloud phone resource interaction process of the cloud phone resource dynamic scheduling method in the specific embodiment provided by the present invention;

[0046] Figure 3 A schematic diagram of a cloud phone resource scheduling process of a cloud phone resource dynamic scheduling method in a specific embodiment provided by the present invention;

[0047] Figure 4 The second flowchart of the method for dynamic scheduling of cloud phone resources provided by the present invention;

[0048] Figure 5 The third flowchart of the method for dynamic scheduling of cloud phone resources provided by the present invention;

[0049] Figure 6 The fourth flowchart of the method for dynamic scheduling of cloud phone resources provided by the present invention;

[0050] Figure 7The fifth flowchart of the method for dynamic scheduling of cloud phone resources provided by the present invention;

[0051] Figure 8 The sixth flow chart of the method for dynamic scheduling of cloud phone resources provided by the present invention;

[0052] Fig. 9 A schematic diagram of the structure of a cloud phone resource dynamic scheduling device provided by the present invention;

[0053] Fig.10 This is a diagram of the internal structure of the computer device provided by the present invention. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments 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.

[0055] Combine the following Figure 1-Figure 10 The present invention describes a method, device, electronic device and storage medium for dynamically scheduling cloud phone resources.

[0056] like Figure 1 As shown, in one embodiment, a cloud phone resource dynamic scheduling method is applied to a dynamic resource management system of a cloud resource center, comprising the following steps:

[0057] Step S110: receiving a first request from a user terminal, where the first request is divided into an instance allocation request and an instance exit request.

[0058] Specifically, the dynamic resource management system receives a request from the user end, that is, a first request, which is divided into an instance allocation request and an instance exit request.

[0059] Step S120, when the first request is an instance allocation request, in response to the instance allocation request, a safety threshold for the number of idle cloud phone instances at each level is formulated according to historical allocation data to determine whether the idle instance pool in the real-time multi-cloud resource cluster has a cloud phone instance that can be safely allocated.

[0060] Specifically, when the first request received in step S110 is an instance allocation request, the dynamic resource management system will respond to the instance allocation request and formulate safety thresholds for the number of idle cloud phone instances at each level based on historical allocation data, and then determine whether the idle instance pool in the real-time multi-cloud resource cluster has cloud phone instances that can be safely allocated.

[0061] Among them, the cloud phone instances that can be safely allocated are cloud phone instance resources whose number of idle cloud phone instances in each level is not less than the safety threshold.

[0062] It should be noted that the safety threshold means that the system is safe only when the number of idle cloud phone instances is higher than the threshold. If the number of idle cloud phone instances is lower than the safety threshold, new instance allocation requests may not be allocated due to insufficient number of idle cloud phone instances, causing system conflicts.

[0063] Step S130, when there is a cloud phone instance that can be safely allocated in the idle instance pool of the real-time multi-cloud resource cluster, based on the instance allocation request and the status of multiple cloud resources in the idle instance pool of the current multi-cloud resource cluster, the optimal instance allocation result is obtained according to the status of multiple cloud resources through a deep reinforcement learning algorithm.

[0064] Specifically, when there are cloud phone instances that can be safely allocated in the idle instance pool of the real-time multi-cloud resource cluster, the dynamic resource management system will obtain the optimal instance allocation result based on the instance allocation request and the status of various cloud resources in the idle instance pool of the current multi-cloud resource cluster through a deep reinforcement learning algorithm according to the allocation strategy corresponding to the status of various cloud resources.

[0065] Among them, the deep reinforcement learning algorithm is used to train the allocation strategies corresponding to different cloud phone instance resources according to the different reward targets corresponding to various cloud resources.

[0066] Step S140, allocates corresponding cloud phone instance resources to the user terminal according to the optimal instance allocation result, and calls the neural network model to extract the user terminal's usage habit characteristics of the cloud phone instance resources according to the instance exit request.

[0067] Specifically, the dynamic resource management system allocates corresponding cloud phone instance resources to the user terminal according to the optimal instance allocation result obtained in step S130, and calls the neural network model to extract the user terminal's usage habit characteristics of the cloud phone instance resources according to the instance exit request.

[0068] Among them, the usage habit characteristics include the user's usage period, usage duration, and resource usage of the allocated cloud phone instance resources.

[0069] Step S150, based on the user's usage habits of the cloud phone instance resources, the cloud phone instance resources are recovered to the idle instance pool according to the allocation source, or the cloud phone instance resources are suspended in the working instance pool.

[0070] Specifically, the dynamic resource management system reclaims the cloud phone instance resources to the idle instance pool according to the allocation source based on the user's usage habits of the cloud phone instance resources, or suspends the cloud phone instance resources in the working instance pool.

[0071] In some embodiments, the cloud phone resource dynamic scheduling method provided by the present invention, when the allocated cloud phone instance resources are recycled to the free instance pool and the corresponding cloud phone instance resources are allocated to the user end according to the optimal instance allocation result, the dynamic resource management system will dynamically update the corresponding cloud resource status in the real-time multi-cloud resource cluster.

[0072] The above-mentioned cloud phone resource dynamic scheduling method uses a neural network model to extract user characteristics and predict the user's usage habits such as the time period, usage duration, and resource usage of the cloud phone. Moreover, the use of the neural network model can accurately predict the user's behavior habits, thereby effectively planning cloud phone resources, and can also make full use of the resources of different cloud platforms to improve resource utilization efficiency, while reducing dependence on a single cloud platform. The resource allocation strategy of the cluster is optimized by using deep reinforcement learning. By analyzing the historical scheduling data in the multi-cloud resource cluster and the various time resource usage patterns of the cloud phone instance, the security threshold of the multi-cloud resource cluster can be predicted according to the real-time situation, and the cloud resource instance can be dynamically adjusted, which improves the intelligence and flexibility of the resource allocation strategy and ensures the safe operation of the system. Therefore, this method can make full use of the resources of different cloud platforms, realize the intelligent management and optimization of cloud phone instance resources, maximize resource utilization efficiency, reduce costs, and provide a good user experience while meeting user needs.

[0073] See also Figure 2 As shown, in a specific embodiment, the present invention provides a method for dynamic scheduling of cloud phone resources. Users request to allocate subscribed cloud phone instances or request to exit the instances in use through their personal mobile phones. Cloud phone instances are divided into instances with different cloud resources such as CPU, memory, network transmission bandwidth, etc. according to the cloud phone level subscribed by the user. The dynamic resource management system identifies different operations based on user requests, and includes two parts: a user behavior habit prediction module and a scheduling strategy module. The user behavior habit prediction module is a neural network prediction model, which predicts the user's habitual time period and habitual duration of using the cloud phone based on the user's historical data on using the cloud phone. The scheduling strategy module is a deep reinforcement learning model, which learns the safe allocation range of the number of idle instances of cloud phones of different levels of multiple cloud resources based on the historical allocation data of multi-cloud resource clusters, and selects from multiple cloud resources (such as Figure 2 Find the optimally scheduled cloud phone instance in the multi-cloud resource cluster composed of "external cloud resource 1, external cloud resource 2, external cloud resource, and a designated operator's cloud resource" and allocate it to the user.

[0074] In this embodiment, if the request sent by the user end is a request for allocating a cloud phone instance, first check the current multi-cloud resource cluster status according to the cloud phone instance information requested by the user, check whether there is a cloud phone instance that can be safely allocated in the idle instance pool in the cluster, and if there is no cloud phone instance that can be safely allocated, wait for a certain period of time and then make a judgment in a loop. If there is a cloud phone instance that can be safely allocated, determine whether it is an instance of the corresponding level. If it is a corresponding level, the corresponding cloud phone instance can be allocated to the user from the corresponding level instances of various cloud resources according to the result given by the scheduling strategy module. If it is a non-corresponding level, the scheduling strategy module can select the best cloud phone instance from the multi-level cloud resource cloud phone instances of multiple levels and allocate it to the user according to the real-time status of the idle instance pool in the multi-cloud resource cluster, and the allocated instance moves from the idle instance pool to the working instance pool.

[0075] In this embodiment, if the request sent by the user is a request to exit the current cloud phone instance, the system will use the user behavior prediction module to predict the user's behavior habits. If it is predicted that the user will not call the cloud phone again in the short term, the corresponding cloud phone instance will be recycled to the corresponding idle instance pool. If it is predicted that the user has a high probability of calling the cloud phone again in the short term, the corresponding current cloud phone instance will be temporarily suspended and still retained in the working instance pool until the user calls it again.

[0076] It should be noted that the multi-cloud resource cluster includes designated operator cloud resources and external cloud resources. The external cloud resources can be one or more public cloud resources from cloud platforms such as Alibaba Cloud and Baidu Cloud. According to different levels of cloud phones, they are divided into instances containing cloud resources such as different standards of CPU, memory, network transmission bandwidth, etc., and are divided into working instance pools in use and idle instance pools waiting to be called. The cloud phone instances in these two instance pools can circulate with each other according to the instance status.

[0077] Combination Figure 3As shown, in one embodiment, the user initiates a request to the dynamic resource management system of the cloud resource center on a personal mobile phone, which is divided into a cloud phone instance allocation request and a current cloud phone instance exit request. If it is an instance allocation request, the policy scheduling module of the dynamic resource management system formulates the safety threshold of the number of idle cloud phone instances at each level according to the historical allocation data, and determines whether there is a cloud phone instance that can be safely allocated in the idle instance pool in the real-time multi-cloud resource cluster. If it cannot be safely allocated, wait for the resources to be safely allocated, and repeat the above steps at intervals. If it can be safely allocated, the scheduling policy module gives the optimal allocation result according to the user request and the status of various cloud resources in the idle instance pool of the current multi-cloud resource cluster. Among them, the policy scheduling module adopts a deep reinforcement learning algorithm to train different strategies according to different reward goals, improve the overall cluster utilization, and can try to allocate cloud phone instances that are geographically close to the user, improve the communication speed between cloud resources and users, and improve the user experience. According to the allocation result given by the scheduling policy module, the optimal cloud phone instance resources are allocated to the user.

[0078] In this embodiment, if the request sent by the user end is a request to exit the current cloud phone instance, the user behavior habit prediction module of the dynamic resource management system predicts the user's usage habits such as the usage period, usage duration, and resource usage. The user behavior habit prediction module uses a neural network model to extract user characteristics, so as to more accurately predict the user's behavior habits and further improve resource utilization efficiency. If it is predicted that the user's behavior habit is that after using the cloud phone during this period, there is a high probability that the cloud phone will not be called again, then the cloud phone instance will be recovered to the original position of the free instance pool according to the source at the time of allocation, waiting for the next allocation. If it is predicted that the user will call it many times after this use, the cloud phone instance will be temporarily suspended and retained in the working instance pool waiting for the user to call it again. In addition, after allocating a cloud phone instance to the user and recycling the cloud phone instance, the dynamic resource management system will promptly update the multi-cloud resource cluster status.

[0079] In this embodiment, the working process and results of the dynamic resource management system are exemplified as follows:

[0080] (1) User 1 subscribes to a cloud phone of level A and requests allocation of cloud phone resources.

[0081] (2) The dynamic resource management system identifies the user request and, based on the user's request for a cloud phone of level A, the scheduling strategy module in the dynamic resource management system determines that it can be safely allocated based on the current state of the multi-cloud resource cluster and gives the optimal instance allocation result.

[0082] (3) According to the optimal allocation result, cloud phone instance 1 with level A in the specified operator's cloud resources is allocated to user 1, and the multi-cloud resource cluster status is updated.

[0083] (4) User 2 subscribes to a cloud phone of level A and requests allocation of cloud phone resources.

[0084] (5) Repeat step (2).

[0085] (6) Based on the optimal allocation result, cloud phone instance 2 with level A in Alibaba Cloud resources is allocated to user 2, and the multi-cloud resource cluster status is updated.

[0086] (7) User 1 applies to exit cloud phone instance 1.

[0087] (8) The dynamic resource management system identifies the user request. The user behavior habit prediction module predicts based on the historical data of user 1 that user 1 is unlikely to call the cloud phone again after this use. Cloud phone instance 1 is recycled to the level A cloud phone instance pool of the designated operator's cloud resource pool, waiting for the next allocation and updating the multi-cloud resource cluster status.

[0088] (9) User 2 applies to exit cloud phone instance 2.

[0089] (10) The dynamic resource management system identifies the user request, and the user behavior prediction module predicts based on the historical data of User 2 that User 2 will call the cloud phone multiple times in a short period of time after this use. The cloud phone instance 2 is temporarily suspended in the working instance pool, waiting for the next call from User 2.

[0090] like Figure 4 As shown, in one embodiment, the cloud phone resource dynamic scheduling method provided by the present invention, when the first request is an instance allocation request, formulates safety thresholds for the number of idle cloud phone instances of each level according to historical allocation data in response to the instance allocation request to determine whether the idle instance pool in the real-time multi-cloud resource cluster has a cloud phone instance that can be safely allocated, and also includes the following steps:

[0091] Step S410, when the first request is an instance exit request, the neural network model is called to predict the user's usage habit characteristics for the current cloud phone instance resources.

[0092] Step S420, based on the user's usage habits of the current cloud phone instance resources, the current cloud phone instance resources are recovered to the corresponding idle instance pool according to the allocation source, or the current cloud phone instance resources are suspended in the working instance pool.

[0093] like Figure 5As shown, in one embodiment, the cloud phone resource dynamic scheduling method provided by the present invention, when the first request is an instance allocation request, formulates a safety threshold value of the number of idle cloud phone instances of each level according to historical allocation data in response to the instance allocation request to determine whether the idle instance pool in the real-time multi-cloud resource cluster has a cloud phone instance that can be safely allocated, specifically comprising the following steps:

[0094] Step S122, judging whether the idle instance pool in the real-time multi-cloud resource cluster has a cloud phone instance that can be safely allocated according to the safety threshold.

[0095] Step S124: in response to the instance allocation request, obtaining an optimal instance allocation result according to a corresponding allocation strategy.

[0096] Specifically, when the judgment result in step S122 is that the idle instance pool in the real-time multi-cloud resource cluster has a cloud phone instance that can be safely allocated, the dynamic resource management system will respond to the instance allocation request and obtain the optimal instance allocation result according to the corresponding allocation strategy.

[0097] Step S126, after the first period of time, it is determined again based on the safety threshold whether there is a cloud phone instance that can be safely allocated in the idle instance pool, so as to wait for the cloud phone instance that can be safely allocated.

[0098] Specifically, when the judgment result in step S122 is that the idle instance pool in the real-time multi-cloud resource cluster does not have a cloud phone instance that can be safely allocated, the dynamic resource management system will re-judge whether there is a cloud phone instance that can be safely allocated in the idle instance pool based on the safety threshold after a set period of time, and wait for a cloud phone instance that can be safely allocated.

[0099] like Figure 6 As shown, in one embodiment, the cloud phone resource dynamic scheduling method provided by the present invention, when there is a cloud phone instance that can be safely allocated in the idle instance pool of the real-time multi-cloud resource cluster, based on the instance allocation request combined with the status of multiple cloud resources in the idle instance pool of the current multi-cloud resource cluster, obtains the optimal instance allocation result according to the status of multiple cloud resources through a deep reinforcement learning algorithm, specifically including the following steps:

[0100] Step S132, obtaining the reward target corresponding to each cloud resource in the real-time multi-cloud resource cluster, and training the allocation strategies corresponding to different reward targets according to each cloud resource and the reward target corresponding to each cloud resource through a deep reinforcement learning algorithm.

[0101] Step S134, calling the trained deep reinforcement learning algorithm to process the states of various cloud resources in the idle instance pool to output optimal allocation results corresponding to the states of the various cloud resources.

[0102] Among them, the optimal allocation result is the cloud phone instance resource with the highest reward target among multiple cloud resources.

[0103] like Figure 7 As shown, in one embodiment, the cloud phone resource dynamic scheduling method provided by the present invention allocates corresponding cloud phone instance resources to the user terminal according to the optimal instance allocation result, and calls the neural network model to extract the user terminal's usage habit characteristics of the cloud phone instance resources according to the instance exit request, specifically including the following steps:

[0104] Step S142, obtain usage habit features of multiple cloud phone instance resources, and use the multiple cloud phone instance resources as input of the neural network model, and use the usage habit features corresponding to each cloud phone instance resource as output of the neural network model to train the neural network model.

[0105] Step S144, based on the instance exit request, calling the trained neural network model to process the allocated cloud phone instance resources to predict the usage habit characteristics corresponding to the allocated cloud phone instance resources.

[0106] like Figure 8 As shown, in one embodiment, the cloud phone resource dynamic scheduling method provided by the present invention recycles the cloud phone instance resources to the idle instance pool according to the allocation source based on the user's usage habit characteristics of the cloud phone instance resources, or suspends the cloud phone instance resources in the working instance pool, specifically including the following steps:

[0107] Step S152, judging whether the cloud phone instance resource is used again within the second time period according to the usage habit characteristics corresponding to the allocated cloud phone instance resource.

[0108] Step S154, suspend the allocated cloud phone instance resources in the working instance pool.

[0109] Specifically, when the judgment result in step S152 is that the cloud phone instance resource is used again within the set time period, the dynamic resource management system will suspend the allocated cloud phone instance resource in the working instance pool.

[0110] Step S156, recovering the allocated cloud phone instance resources to the original idle instance pool according to the allocation source.

[0111] Specifically, when the judgment result in step S152 is that the cloud phone instance resources will not be used again within the set time period, the dynamic resource management system will recycle the allocated cloud phone instance resources to the original idle instance pool according to the allocation source.

[0112] The cloud phone resource dynamic scheduling device provided by the present invention is described below. The cloud phone resource dynamic scheduling device described below and the cloud phone resource dynamic scheduling method described above can be referenced to each other.

[0113] like Fig. 9 As shown, in one embodiment, a cloud phone resource dynamic scheduling device includes a request receiving module 910, a request response module 920, a resource allocation module 930, a user behavior prediction module 940 and a resource processing module 950.

[0114] The request receiving module 910 is used to receive a first request from a user terminal, where the first request is divided into an instance allocation request and an instance exit request.

[0115] The request response module 920 is used to formulate safety thresholds for the number of idle cloud phone instances at various levels based on historical allocation data in response to the instance allocation request when the first request is an instance allocation request, so as to determine whether the idle instance pool in the real-time multi-cloud resource cluster has cloud phone instances that can be safely allocated.

[0116] The resource allocation module 930 is used to obtain the optimal instance allocation result based on the status of multiple cloud resources in the idle instance pool of the real-time multi-cloud resource cluster based on the instance allocation request when there is a cloud phone instance that can be safely allocated in the idle instance pool of the real-time multi-cloud resource cluster, through a deep reinforcement learning algorithm.

[0117] The user behavior prediction module 940 is used to allocate corresponding cloud phone instance resources to the user terminal according to the optimal instance allocation result, and call the neural network model to extract the user terminal's usage habit characteristics of the cloud phone instance resources according to the instance exit request.

[0118] The resource processing module 950 is used to recycle the cloud phone instance resources to the idle instance pool according to the allocation source based on the user's usage habits of the cloud phone instance resources, or to suspend the cloud phone instance resources in the working instance pool.

[0119] Among them, the cloud phone instances that can be safely allocated are cloud phone instance resources in each level whose number of idle cloud phone instances is not less than the safety threshold. The deep reinforcement learning algorithm is used to train allocation strategies corresponding to different cloud phone instance resources according to different reward targets corresponding to multiple cloud resources. The usage habit characteristics include the user's use period, usage duration and resource usage of the allocated cloud phone instance resources.

[0120] In this embodiment, the cloud phone resource dynamic scheduling device provided by the present invention further includes an instance exit request response module, which is used to:

[0121] When the first request is an instance exit request, the neural network model is called to predict the user's usage habit characteristics for the current cloud phone instance resources.

[0122] Based on the user's usage habits of the current cloud phone instance resources, the current cloud phone instance resources are recovered to the corresponding idle instance pool according to the allocation source, or the current cloud phone instance resources are suspended in the working instance pool.

[0123] In this embodiment, the cloud phone resource dynamic scheduling device provided by the present invention, the request response module 920 is specifically used for:

[0124] According to the security threshold, determine whether the idle instance pool in the real-time multi-cloud resource cluster has a cloud phone instance that can be safely allocated. If so,

[0125] In response to the instance allocation request, obtain the optimal instance allocation result according to the corresponding allocation strategy.

[0126] After the first period of time, it is determined again based on the safety threshold whether there is a cloud phone instance that can be safely allocated in the idle instance pool, so as to wait for the cloud phone instance that can be safely allocated.

[0127] In this embodiment, the resource allocation module 930 of the cloud phone resource dynamic scheduling device provided by the present invention is specifically used for:

[0128] The reward target corresponding to each cloud resource in the real-time multi-cloud resource cluster is obtained, and the allocation strategies corresponding to different reward targets are trained according to each cloud resource and the reward target corresponding to each cloud resource through the deep reinforcement learning algorithm.

[0129] The trained deep reinforcement learning algorithm is called to process the status of various cloud resources in the idle instance pool to output the optimal allocation results corresponding to the status of various cloud resources.

[0130] Among them, the optimal allocation result is the cloud phone instance resource with the highest reward target among multiple cloud resources.

[0131] In this embodiment, the user behavior prediction module 940 of the cloud phone resource dynamic scheduling device provided by the present invention is specifically used for:

[0132] The usage habit features of multiple cloud phone instance resources are obtained, and the multiple cloud phone instance resources are used as the input of the neural network model. The usage habit features corresponding to each cloud phone instance resource are used as the output of the neural network model to train the neural network model.

[0133] Based on the instance exit request, the trained neural network model is called to process the allocated cloud phone instance resources to predict the usage habit characteristics corresponding to the allocated cloud phone instance resources.

[0134] In this embodiment, the resource processing module 950 of the cloud phone resource dynamic scheduling device provided by the present invention is specifically used for:

[0135] According to the usage habit characteristics corresponding to the allocated cloud phone instance resources, determine whether the cloud phone instance resources are used again within the second time period. If so, then

[0136] Suspend the allocated cloud phone instance resources in the working instance pool. If not, then

[0137] The allocated cloud phone instance resources are recovered to the original idle instance pool according to the allocation source.

[0138] In this embodiment, the cloud phone resource dynamic scheduling device provided by the present invention further includes a resource update module for:

[0139] When the allocated cloud phone instance resources are recycled to the idle instance pool and the corresponding cloud phone instance resources are allocated to the user end according to the optimal instance allocation result, the corresponding cloud resource status in the real-time multi-cloud resource cluster is dynamically updated.

[0140] Fig.10 An example of a physical structure diagram of an electronic device is shown. The electronic device may be a smart terminal, and its internal structure diagram may be as follows: Fig.10 As shown. The electronic device includes a processor, a memory and a network interface connected via a system bus. Among them, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the electronic device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for dynamically scheduling cloud phone resources is implemented, and the method includes:

[0141] receiving a first request from a user terminal, the first request being divided into an instance allocation request and an instance exit request;

[0142] When the first request is an instance allocation request, in response to the instance allocation request, a safety threshold for the number of idle cloud phone instances at each level is formulated according to historical allocation data to determine whether an idle instance pool in a real-time multi-cloud resource cluster has a cloud phone instance that can be safely allocated;

[0143] When there is a cloud phone instance that can be safely allocated in the idle instance pool of the real-time multi-cloud resource cluster, based on the instance allocation request and the status of multiple cloud resources in the idle instance pool of the current multi-cloud resource cluster, the optimal instance allocation result is obtained according to the status of multiple cloud resources through a deep reinforcement learning algorithm;

[0144] Allocate corresponding cloud phone instance resources to the user according to the optimal instance allocation result, and call the neural network model to extract the user's usage habit characteristics of cloud phone instance resources according to the instance exit request;

[0145] Based on the user's usage habits of cloud phone instance resources, the cloud phone instance resources are recycled to the idle instance pool according to the allocation source, or the cloud phone instance resources are suspended in the working instance pool;

[0146] Among them, the cloud phone instances that can be safely allocated are cloud phone instance resources in each level whose number of idle cloud phone instances is not less than the safety threshold. The deep reinforcement learning algorithm is used to train allocation strategies corresponding to different cloud phone instance resources according to different reward targets corresponding to multiple cloud resources. The usage habit characteristics include the user's use period, usage duration and resource usage of the allocated cloud phone instance resources.

[0147] Those skilled in the art will understand that Fig.10 The structure shown in the figure is only a block diagram of a part of the structure related to the scheme of the present invention, and does not constitute a limitation on the electronic device to which the scheme of the present invention is applied. The specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0148] On the other hand, the present invention also provides a computer storage medium storing a computer program, which implements a method for dynamically scheduling cloud phone resources when the computer program is executed by a processor, the method comprising:

[0149] receiving a first request from a user terminal, the first request being divided into an instance allocation request and an instance exit request;

[0150] When the first request is an instance allocation request, in response to the instance allocation request, a safety threshold for the number of idle cloud phone instances at each level is formulated according to historical allocation data to determine whether an idle instance pool in a real-time multi-cloud resource cluster has a cloud phone instance that can be safely allocated;

[0151] When there is a cloud phone instance that can be safely allocated in the idle instance pool of the real-time multi-cloud resource cluster, based on the instance allocation request and the status of multiple cloud resources in the idle instance pool of the current multi-cloud resource cluster, the optimal instance allocation result is obtained according to the status of multiple cloud resources through a deep reinforcement learning algorithm;

[0152] Allocate corresponding cloud phone instance resources to the user according to the optimal instance allocation result, and call the neural network model to extract the user's usage habit characteristics of cloud phone instance resources according to the instance exit request;

[0153] Based on the user's usage habits of cloud phone instance resources, the cloud phone instance resources are recycled to the idle instance pool according to the allocation source, or the cloud phone instance resources are suspended in the working instance pool;

[0154] Among them, the cloud phone instances that can be safely allocated are cloud phone instance resources in each level whose number of idle cloud phone instances is not less than the safety threshold. The deep reinforcement learning algorithm is used to train allocation strategies corresponding to different cloud phone instance resources according to different reward targets corresponding to multiple cloud resources. The usage habit characteristics include the user's use period, usage duration and resource usage of the allocated cloud phone instance resources.

[0155] In another aspect, a computer program product or a computer program is provided, the computer program product or the computer program comprising computer instructions, the computer instructions being stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium, and when the processor executes the computer instructions, a method for dynamically scheduling cloud phone resources is implemented, the method comprising:

[0156] receiving a first request from a user terminal, the first request being divided into an instance allocation request and an instance exit request;

[0157] When the first request is an instance allocation request, in response to the instance allocation request, a safety threshold for the number of idle cloud phone instances at each level is formulated according to historical allocation data to determine whether an idle instance pool in a real-time multi-cloud resource cluster has a cloud phone instance that can be safely allocated;

[0158] When there is a cloud phone instance that can be safely allocated in the idle instance pool of the real-time multi-cloud resource cluster, based on the instance allocation request and the status of multiple cloud resources in the idle instance pool of the current multi-cloud resource cluster, the optimal instance allocation result is obtained according to the status of multiple cloud resources through a deep reinforcement learning algorithm;

[0159] Allocate corresponding cloud phone instance resources to the user according to the optimal instance allocation result, and call the neural network model to extract the user's usage habit characteristics of cloud phone instance resources according to the instance exit request;

[0160] Based on the user's usage habits of cloud phone instance resources, the cloud phone instance resources are recycled to the idle instance pool according to the allocation source, or the cloud phone instance resources are suspended in the working instance pool;

[0161] Among them, the cloud phone instances that can be safely allocated are cloud phone instance resources in each level whose number of idle cloud phone instances is not less than the safety threshold. The deep reinforcement learning algorithm is used to train allocation strategies corresponding to different cloud phone instance resources according to different reward targets corresponding to multiple cloud resources. The usage habit characteristics include the user's use period, usage duration and resource usage of the allocated cloud phone instance resources.

[0162] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing related hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory.

[0163] By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0164] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0165] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A method for dynamically scheduling cloud phone resources, characterized in that: A dynamic resource management system applied to a cloud resource center, the method comprising: receiving a first request from a user terminal, wherein the first request is divided into an instance allocation request and an instance exit request; When the first request is an instance allocation request, in response to the instance allocation request, a safety threshold for the number of idle cloud phone instances at each level is formulated according to historical allocation data to determine whether an idle instance pool in a real-time multi-cloud resource cluster has a cloud phone instance that can be safely allocated; When there is a cloud phone instance that can be safely allocated in the idle instance pool of the real-time multi-cloud resource cluster, based on the instance allocation request and the status of multiple cloud resources in the idle instance pool of the current multi-cloud resource cluster, an optimal instance allocation result is obtained according to the status of the multiple cloud resources through a deep reinforcement learning algorithm; Allocate corresponding cloud phone instance resources to the user terminal according to the optimal instance allocation result, and call a neural network model to extract the user terminal's usage habit characteristics of the cloud phone instance resources according to the instance exit request; Based on the usage habit characteristics of the cloud phone instance resources by the user end, the cloud phone instance resources are recycled to the idle instance pool according to the allocation source, or the cloud phone instance resources are suspended in the working instance pool; Among them, the safely allocated cloud phone instances are cloud phone instance resources in which the number of idle cloud phone instances in each level is not less than the safety threshold, the deep reinforcement learning algorithm is used to train allocation strategies corresponding to different cloud phone instance resources according to different reward targets corresponding to the multiple cloud resources, and the usage habit characteristics include the user's use time period, usage duration and resource usage of the allocated cloud phone instance resources.

2. The cloud phone resource dynamic scheduling method according to claim 1 is characterized in that: When the first request is an instance allocation request, in response to the instance allocation request, a safety threshold for the number of idle cloud phone instances at each level is formulated according to historical allocation data to determine whether an idle instance pool in a real-time multi-cloud resource cluster has a cloud phone instance that can be safely allocated, and the above also includes: When the first request is an instance exit request, calling the neural network model to predict the user's usage habit characteristics for the current cloud phone instance resources; Based on the user's usage habits of the current cloud phone instance resources, the current cloud phone instance resources are recovered to the corresponding idle instance pool according to the allocation source, or the current cloud phone instance resources are suspended in the working instance pool.

3. The cloud phone resource dynamic scheduling method according to claim 2 is characterized in that: When the first request is an instance allocation request, formulating safety thresholds for the number of idle cloud phone instances at various levels according to historical allocation data in response to the instance allocation request to determine whether an idle instance pool in a real-time multi-cloud resource cluster has a cloud phone instance that can be safely allocated, including: According to the safety threshold, it is determined whether the idle instance pool in the real-time multi-cloud resource cluster has a cloud phone instance that can be safely allocated; if so, In response to the instance allocation request, obtain the optimal instance allocation result according to the corresponding allocation strategy; if not, then After the first period of time, it is determined again based on the safety threshold whether there is a cloud phone instance that can be safely allocated in the idle instance pool, so as to wait for the cloud phone instance that can be safely allocated.

4. The cloud phone resource dynamic scheduling method according to claim 1 is characterized in that: When there is a cloud phone instance that can be safely allocated in the idle instance pool of the real-time multi-cloud resource cluster, based on the instance allocation request and the status of multiple cloud resources in the idle instance pool of the current multi-cloud resource cluster, an optimal instance allocation result is obtained according to the status of the multiple cloud resources through a deep reinforcement learning algorithm, including: Obtaining a reward target corresponding to each cloud resource in the real-time multi-cloud resource cluster, and training allocation strategies corresponding to different reward targets according to each cloud resource and the reward target corresponding to each cloud resource through the deep reinforcement learning algorithm; Calling the trained deep reinforcement learning algorithm to process the states of the various cloud resources in the idle instance pool to output optimal allocation results corresponding to the states of the various cloud resources; Among them, the optimal allocation result is the cloud phone instance resource with the highest reward target among the multiple cloud resources.

5. The cloud phone resource dynamic scheduling method according to claim 1 is characterized in that: The allocating corresponding cloud phone instance resources to the user terminal according to the optimal instance allocation result, and calling a neural network model to extract the user terminal's usage habit characteristics of the cloud phone instance resources according to the instance exit request, includes: Obtaining usage habit features of multiple cloud phone instance resources, and using the multiple cloud phone instance resources as inputs of the neural network model, using the usage habit features corresponding to each cloud phone instance resource as outputs of the neural network model, and training the neural network model; Based on the instance exit request, the trained neural network model is called to process the allocated cloud phone instance resources to predict the usage habit characteristics corresponding to the allocated cloud phone instance resources.

6. The method for dynamically scheduling cloud phone resources according to claim 5, characterized in that: The recovering the cloud phone instance resources to the idle instance pool according to the allocation source based on the usage habit characteristics of the cloud phone instance resources by the user end, or suspending the cloud phone instance resources in the working instance pool, includes: Determine whether the cloud phone instance resource is used again within the second time period according to the usage habit characteristics corresponding to the allocated cloud phone instance resource; if so, Suspend the allocated cloud phone instance resources in the working instance pool; if not, The allocated cloud phone instance resources are recovered to the original idle instance pool according to the allocation source.

7. The method for dynamically scheduling cloud phone resources according to any one of claims 1 to 6, characterized in that: The method further comprises: When the allocated cloud phone instance resources are recycled to the idle instance pool and the corresponding cloud phone instance resources are allocated to the user terminal according to the optimal instance allocation result, the corresponding cloud resource status in the real-time multi-cloud resource cluster is dynamically updated.

8. A cloud phone resource dynamic scheduling device, characterized in that: A dynamic resource management system applied to a cloud resource center, the device comprising: A request receiving module, used to receive a first request from a user terminal, wherein the first request is divided into an instance allocation request and an instance exit request; A request response module, for, when the first request is an instance allocation request, formulating safety thresholds for the number of idle cloud phone instances at various levels according to historical allocation data in response to the instance allocation request, so as to determine whether an idle instance pool in a real-time multi-cloud resource cluster has a cloud phone instance that can be safely allocated; A resource allocation module, configured to obtain an optimal instance allocation result according to the status of multiple cloud resources by using a deep reinforcement learning algorithm based on the instance allocation request and the status of multiple cloud resources in the idle instance pool of the current multi-cloud resource cluster when there is a cloud phone instance that can be safely allocated in the idle instance pool of the real-time multi-cloud resource cluster; A user behavior prediction module is used to allocate corresponding cloud phone instance resources to the user terminal according to the optimal instance allocation result, and call a neural network model to extract the user terminal's usage habit characteristics of the cloud phone instance resources according to the instance exit request; A resource processing module, configured to recycle the cloud phone instance resources to the idle instance pool according to the allocation source, or suspend the cloud phone instance resources in the working instance pool based on the usage habit characteristics of the cloud phone instance resources by the user end; Among them, the safely allocated cloud phone instances are cloud phone instance resources in which the number of idle cloud phone instances in each level is not less than the safety threshold, the deep reinforcement learning algorithm is used to train allocation strategies corresponding to different cloud phone instance resources according to different reward targets corresponding to the multiple cloud resources, and the usage habit characteristics include the user's use time period, usage duration and resource usage of the allocated cloud phone instance resources.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images