Communication service providing method and apparatus, base station, server, and storage medium
By deploying machine learning models on base stations to provide inference services for communication services, the problem of high human involvement in network operation and maintenance has been solved, efficiency has been improved and costs have been reduced, and efficient network optimization has been achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ZTE CORP
- Filing Date
- 2019-09-26
- Publication Date
- 2026-07-07
AI Technical Summary
High levels of human involvement in network operations and maintenance lead to low efficiency and increased costs in network optimization.
Deploying machine learning models on base stations allows for the provision of inference services for communication services, including site health checks, fault prediction, base station planning, and radio frequency planning, replacing manual processing.
It improved network operating efficiency, saved costs, and reduced latency and enhanced data security through direct inference.
Smart Images
Figure CN112561070B_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present invention relate to, but are not limited to, the field of communications, and specifically, to, but are not limited to, methods, apparatuses, base stations, servers, and storage media for providing communication services. Background Technology
[0002] In the CT (Communication Technology) field, with the explosive growth of business volume, communication networks face huge business volume and massive data load, thus network operation and maintenance face enormous challenges and low efficiency.
[0003] Meanwhile, the high degree of human involvement in network operations and maintenance not only reduces network optimization efficiency but also increases network operations and maintenance costs. Summary of the Invention
[0004] The communication service provision method, apparatus, base station, server, and storage medium provided in this invention mainly address the technical problem that, in related technologies, network operation and maintenance involve a high degree of human intervention, which reduces network optimization efficiency and increases network operation and maintenance costs.
[0005] To address the aforementioned technical problems, embodiments of the present invention provide a method for providing communication services, comprising: a base station providing inference services for communication services through a machine learning model deployed on itself.
[0006] This invention also provides a communication service providing apparatus, which includes: a service module for providing communication services through a machine learning model.
[0007] This invention also provides a machine learning model providing apparatus, comprising: a storage module for storing machine learning models for use by a communication service providing apparatus to retrieve the machine learning models, the machine learning models being used to provide services for communication services.
[0008] This invention also provides a base station, which includes: a first processor, a first memory, and a first communication bus;
[0009] The first communication bus is used to realize the connection and communication between the first processor and the first memory;
[0010] The first processor is used to execute one or more programs stored in the first memory to implement the steps on the base station side of the communication service provision method described above.
[0011] This invention also provides a server, which includes: a second processor, a second memory, and a second communication bus;
[0012] The second communication bus is used to realize the connection and communication between the second processor and the second memory;
[0013] The second processor is used to execute one or more programs stored in the second memory to implement the steps on the OMC side of the above-described communication service provision method.
[0014] This invention also provides a storage medium storing one or more programs, which can be executed by one or more processors to implement the steps on the base station side or OMC side in the above-described communication service provision method.
[0015] The beneficial effects of this invention are:
[0016] According to the communication service provision method, apparatus, base station, server and storage medium provided in the embodiments of the present invention, the base station provides inference services for communication services through a machine learning model deployed on itself. In some implementations, since the base station provides inference services for communication services through the machine learning model, the efficiency of network operation is improved due to the advantages of the machine learning model itself in data processing. Furthermore, the application of the machine learning model replaces manual processing in some aspects, saving costs.
[0017] Other features and corresponding beneficial effects of the present invention will be described in the latter part of the specification, and it should be understood that at least some of the beneficial effects will become obvious from the description in the specification. Attached Figure Description
[0018] Figure 1 A flowchart of a method for providing communication service according to Embodiment 1 of the present invention is provided.
[0019] Figure 2 This is a flowchart of the model download process according to Embodiment 1 of the present invention;
[0020] Figure 3 This is a flowchart illustrating the OMC's control process for the base station according to Embodiment 1 of the present invention.
[0021] Figure 4 This is a flowchart of the model rollback process in Embodiment 1 of the present invention;
[0022] Figure 5 This is a schematic diagram of the interaction between the OMC and the base station during the model fallback process in Embodiment 1 of the present invention;
[0023] Figure 6 This is a flowchart of the model version upgrade process according to Embodiment 1 of the present invention;
[0024] Figure 7 This is a schematic diagram illustrating the interaction between the OMC and the base station during the model version upgrade process in Embodiment 1 of the present invention.
[0025] Figure 8 This is a system architecture diagram of an optional Serving SDK according to Embodiment 1 of the present invention;
[0026] Figure 9 A flowchart illustrating the method for providing communication service according to Embodiment 2 of the present invention;
[0027] Figure 10 This is a schematic diagram of multi-model collaboration in Embodiment 2 of the present invention;
[0028] Figure 11 This is a system architecture diagram of an optional Serving system according to Embodiment 2 of the present invention;
[0029] Figure 12 This is a schematic diagram of the base station structure according to Embodiment 4 of the present invention;
[0030] Figure 13 This is a schematic diagram of the server structure according to Embodiment 4 of the present invention. Detailed Implementation
[0031] To make the objectives, technical solutions, and advantages of this invention clearer, the embodiments of this invention will be further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
[0032] Example 1:
[0033] To address the problems of high network operation and maintenance costs and low efficiency in existing networks, this invention proposes a method for providing communication services. Please refer to [link to relevant documentation]. Figure 1 ,include:
[0034] S101. The base station provides inference services for communication services through machine learning models deployed on its own.
[0035] In this embodiment of the invention, a machine learning model is deployed on the base station, which then provides inference services for communication services through the machine learning model. That is, in this embodiment, the machine learning model is applied to communication aspects, such as network planning and optimization. The base station can provide at least one of the following services for communication services through the machine learning model: site health check service, fault prediction service, base station planning service, radio frequency planning service, traffic prediction service, load prediction service, KPI (Key Performance Indicator) prediction service, coverage optimization service, parameter optimization service, and equipment energy saving service. In other words, in this embodiment, the machine learning model can be applied to site health checks, fault prediction, base station planning, radio frequency planning, traffic prediction, load prediction, KPI prediction, coverage optimization, parameter optimization, and equipment energy saving. The machine learning model can provide REST or RPC type inference services.
[0036] In this embodiment of the invention, the machine learning model can be a machine learning model designed based on at least one machine learning model running framework among Keras, TensorFlow, PyTorch, Caffe, and MxNet.
[0037] In this embodiment of the invention, before the base station provides inference services for communication services through the machine learning model deployed on its own platform, the base station also needs to download and deploy the machine learning model. When deploying the machine learning model, the base station can deploy it in an independent microservice. Alternatively, since the computing resources on the base station are limited and more stringent in terms of latency and power consumption, to maximize performance, the machine learning model and communication services can be deployed in the same microservice. In this way, the communication services can directly implement the model's inference requests through function calls.
[0038] In this embodiment of the invention, the OMC (Operation and Maintenance Center) stores trained machine learning models, which the base station can download. In other words, in this embodiment, the offline-trained machine learning model is uploaded to the OMC for the base station to download. It should be understood that in this embodiment, based on the needs of communication services, the network structure of the machine learning model can be designed using at least one of the following machine learning model runtime frameworks: Keras, TensorFlow, PyTorch, Caffe, and MxNet. The training dataset can then be accessed to complete the training of the machine learning model. The trained machine learning model can be more easily exported to a compilation and optimization toolchain. After model optimization such as model pruning, layer fusion, and accuracy reduction, an optimized binary machine learning model is released and uploaded to the OMC. The OMC can store the machine learning model in a MODEL STORE.
[0039] It is understandable that different base stations have different hardware environments and support different runtime environments for machine learning models. The runtime environment for a machine learning model includes, but is not limited to, the hardware runtime environment and / or the machine learning model runtime framework. The hardware runtime environment can be hardware such as a GPU (Graphics Processing Unit) or a CPU (Central Processing Unit). For example, some base stations only include a CPU, so machine learning models that only support GPU runtimes cannot run on that base station. To ensure that different base stations can download machine learning models that match their supported runtimes, in this embodiment of the invention, for the same machine learning model, at least two types of machine learning models can be published in the OMC. Different types of machine learning models support different runtime environments; that is, if two machine learning models have different runtime frameworks, their types are different; if two machine learning models support different hardware runtime environments, their types are different; if two machine learning models have different runtime frameworks and different supported hardware runtime environments, their types are different. In this way, base stations can download machine learning models that match their supported runtimes from at least two types of machine learning models. For example, a base station can download an image package that matches its own hardware environment from at least two types of image packages.
[0040] For example, for the same machine learning model, the following three types of machine learning models can be published in OMC:
[0041] Machine learning model 1: Supports CPU and TensorFlow runtime;
[0042] Machine learning model 2: Supports CPU and OpenVINO runtime;
[0043] Machine learning model 3: Supports GPU and TensorRT runtime.
[0044] For a given base station, assuming it only includes a CPU, its supported runtime is as follows:
[0045] -serving-tf-cpu:1.0: Supports CPU, TensorFlow runtime;
[0046] -serving-openvino-cpu:2.0: Supports CPU, OpenVINO runtime;
[0047] -serving-tf-openvino-cpu:1.0: Supports CPU, TensorFlow, or OpenVINO runtime.
[0048] Therefore, the base station can download machine learning model one and / or machine learning model two that match its supported runtime.
[0049] For example, see below, the OMC MODEL STORE released a model called resnet50, which has two versions, and within each version, two different models are deployed for different hardware architectures.
[0050]
[0051] The config.pbtxt file contains configuration information about model metadata, including model inputs and outputs, inference hyperparameter configurations, and model configurations corresponding to the hardware architecture. See below for an example of a config.pbtxt file.
[0052]
[0053] The process by which the base station downloads the machine learning model from the OMC can be found in [reference needed]. Figure 2 As shown, it includes:
[0054] S201 and OMC send model download instructions to the base station.
[0055] OMC can send model download instructions to the base stations it manages.
[0056] S202, The base station downloads the machine learning model from the OMC.
[0057] After receiving the model download instruction from the OMC, the base station downloads the machine learning model from the OMC using at least one of the following protocols: SSH (Secure Shell), HTTPS (Hypertext Transfer Protocol Secure), or FTPS (File Transfer Protocol). If the download is successful, the base station notifies the OMC; if it fails, the OMC attempts to resume the download process.
[0058] After downloading the machine learning model, the base station needs to activate and load the model to complete its deployment. Activation means enabling the machine learning model, and loading means deserializing the model to achieve this. Furthermore, if a new version of the machine learning model becomes available after deployment, the base station can upgrade its own model to that version. Alternatively, if a rollback to a specific version is needed after deployment, the base station can revert to that version.
[0059] In this embodiment of the invention, the activation, loading, version upgrade, and version rollback of the machine learning model on the base station can be controlled by the OMC. For example, see... Figure 3 As shown, the OMC's control process for the machine learning model on the base station includes:
[0060] S301 and OMC send model control commands to the base station.
[0061] In this embodiment of the invention, the model control instructions include, but are not limited to, at least one of the following control instructions: activation instruction, loading instruction, version upgrade instruction, and rollback instruction.
[0062] S302. The base station controls the machine learning model according to the model control instructions.
[0063] After receiving the model control command sent by the OMC, the base station controls the machine learning model according to the model control command. The model control command includes at least one of the following: activation command, loading command, version upgrade command, and rollback command. The base station activates the machine learning model according to the activation command, loads the machine learning model according to the loading command, upgrades the version of the machine learning model according to the version upgrade command, and rolls back the version of the machine learning model to a specified version according to the rollback command.
[0064] For example, see Figure 4 As shown, the fallback process of a machine learning model includes:
[0065] S401 and OMC send a rollback command to the base station.
[0066] In this embodiment of the invention, the rollback instruction may include a rollback strategy, which includes the version number of the historical version to which the rollback should be performed.
[0067] S402. The base station rolls back the version of the machine learning model to the specified version according to the rollback command.
[0068] In this embodiment of the invention, after receiving a rollback instruction, the base station unloads the current version of the machine learning model in the base station. At the same time, it automatically loads the historical version of the machine learning model corresponding to the version number in the rollback strategy, deserializes the historical version of the machine learning model, and constructs an inference runtime context, so that the historical version of the machine learning model can provide REST type or RPC type inference services.
[0069] See Figure 5 As shown, this section uses the example of reverting from version V2 to version V1 to illustrate the process, including:
[0070] S501 and OMC send RollbackModel(V1) to the base station.
[0071] OMC sends model rollback instructions, including version V1, to the base station.
[0072] S502, Base Station OnBoarding V1 & Stop V2.
[0073] The base station loads the V1 version of the machine learning model and stops loading the V2 version of the machine learning model.
[0074] S503, Base Station Servable V1.
[0075] The base station provides inference services through the V1 version of the machine learning model.
[0076] See Figure 6 As shown, the version upgrade process of a machine learning model includes:
[0077] S601 and OMC send version upgrade instructions to the base station.
[0078] In this embodiment of the invention, when there is a new version of the machine learning model, the OMC can send a version upgrade instruction to the base station. The version upgrade instruction may include a version upgrade control strategy, such as the version number to be upgraded.
[0079] S602. The base station upgrades the version of the machine learning model according to the version upgrade instruction.
[0080] After receiving the version upgrade instruction, the base station downloads the new version of the machine learning model, activates and loads the new version of the machine learning model, and uninstalls the old version of the machine learning model.
[0081] join Figure 7 As shown, this example illustrates the upgrade from version V1 to version V2.
[0082] S701 and OMC send UpgradeModel(V2) to the base station.
[0083] OMC sends an instruction to the base station to download the V2 version of the machine learning model.
[0084] S702, Base Station OnBoarding V2 & Stop V1.
[0085] The base station loads the V2 version of the machine learning model and stops the V1 version of the machine learning model.
[0086] S703, Base Station Servable V2.
[0087] The base station provides inference services through the V2 version of the machine learning model.
[0088] In this embodiment of the invention, the base station can also poll the machine learning models in the OMC to automatically discover new versions of the machine learning models. For example, when the machine learning model is stored in the MODEL STORE, the base station can discover new versions of the machine learning model by polling the MODEL STORE in the OMC. After discovering a new version of the machine learning model, the base station can decide when to download, activate, and load the new version of the machine learning model based on its own service situation, network status, etc.
[0089] In this embodiment of the invention, a Serving SDK (Software Development Kit) for machine learning models can be configured on the base station to manage the machine learning models deployed on the base station. The Serving SDK includes at least one of the following functions: model uploading, model upgrading, model scheduling, model inference, model monitoring, and runtime isolation.
[0090] In this embodiment of the invention, the Serving SDK includes an API (Application Programming Interface) for receiving control information for machine learning models and providing corresponding feedback based on the control information. In other words, users can control the machine learning models on the base station through the Serving SDK's API. The API can be a C-based API or a C++-based API. In this embodiment, the control information can be a query command, meaning users can use the API to query the machine learning models deployed on the base station, their running status, etc. The control information can also be a customized solution for the machine learning model; that is, the API can be used to extend customized solutions for the machine learning model. These customized solutions include, but are not limited to, extensions to the functionality of the machine learning model, multi-model collaboration strategies, runtime, etc. In other words, in this embodiment, users can develop and customize machine learning models through the API.
[0091] For example, see Figure 8 As shown, Figure 8 This is a system architecture diagram of the Serving SDK. The Serving SDK includes a C++ API for receiving user control information, and also includes model management (ModelManager), model scheduling (Scheduler), and ModelStore (model storage) functions. The Serving SDK also encapsulates different runtimes, such as the TensorFlow runtime for CPU or GPU; the OpenVINO runtime for CPU; the ML runtime for CPU; and device and custom runtimes. The device and custom runtimes can be configured by the user, meaning the user can implement a custom inference runtime environment based on the specific hardware architecture. In this embodiment of the invention, the Serving (microservice, e.g., a microservice deploying a machine learning model) in the base station can manage the machine learning model by calling the Serving SDK. Of course, in other examples, the Caffe or PyTorch runtime environment can also be selected according to the actual situation.
[0092] It should be noted that, in this embodiment of the invention, a base station may deploy one machine learning model or at least two machine learning models. When at least two machine learning models are deployed on a base station, the machine learning models on the base station can independently complete inference services, or they can collaborate with each other to complete inference services. Specifically, the base station can call at least two machine learning models from the at least two machine learning models deployed on the base station for collaboration according to a multi-model collaboration strategy to provide inference services for communication services.
[0093] In this embodiment of the invention, a multi-model collaboration strategy can be set within the machine learning model. That is, the multi-model collaboration strategy is set within the algorithm of the machine learning model. During the execution of the machine learning model, related machine learning models are automatically invoked for collaboration based on this strategy to complete the inference service. For example, in a TensorFlow-based machine learning model, a control operation (OP) is used to set the multi-model collaboration strategy within the algorithm. This can be achieved through operations like Switch, Merge, Enter, Leave, and Next, similar to if-else, while, and for statements in high-level languages. During training, each machine learning model can be trained independently. During deployment, the machine learning models, including the control operation, can be deserialized. During inference, the machine learning models, including the control operation, execute the control operation, thereby invoking other related machine learning models for collaborative inference.
[0094] Alternatively, the base station can receive multi-model collaboration strategies via the API in the Serving SDK. This means users can control at least two machine learning models on the base station to collaborate through the API interface. These multi-model collaboration strategies can be imperative, such as using native C++ to edit the strategy and control the collaboration of multiple machine learning models; or they can be declarative, using the DSL (Domain Specified Language) provided by the Serving SDK to construct a multi-model orchestrator, thereby invoking at least two machine learning models to collaborate. The most basic unit of DSL scheduling is a ScheduleAction. Each ScheduleAction can represent data preprocessing or inference of a sub-model. Collaboration methods include, but are not limited to, sequential execution, branch execution, and iterative execution.
[0095] The communication service provision method provided in this embodiment of the invention allows a base station to provide inference services for communication services through a machine learning model deployed on its own. In some implementations, because the base station provides inference services for communication services through the machine learning model, the advantages of the machine learning model itself in data processing improve the efficiency of network operation. At the same time, the application of the machine learning model replaces manual processing in some aspects, saving costs. Furthermore, the base station can directly perform inference based on the machine learning model after receiving mobile phone data and complete the corresponding decisions. Compared with the scheme where the base station sends the collected data to a higher-level network (such as OMC) for decision-making, it can not only reduce latency but also improve data security.
[0096] Example 2:
[0097] To better understand this invention, embodiments are described in conjunction with more specific examples. See also Figure 9 As shown, Figure 9 A flowchart of a method for providing communication service according to embodiments of the present invention includes:
[0098] S901 and OMC send model download instructions to the base station.
[0099] In this embodiment of the invention, the OMC's MODEL STORE stores machine learning models trained based on various communication service requirements. Each machine learning model is released in binary form. For the same machine learning model, there are at least two types of machine learning models. Different types of machine learning models support different runtime environments. The runtime environment includes the machine learning model's runtime framework and the hardware runtime environment. It should be noted that the machine learning model's runtime framework includes, but is not limited to, at least one of the following: Keras, TensorFlow, PyTorch, Caffe, and MxNet. The hardware runtime environment includes, but is not limited to, GPUs and CPUs.
[0100] For example, for the same machine learning model, the following four types of machine learning models can be published in OMC:
[0101] Machine Learning Model 1.0: Supports CPU and TensorFlow runtime;
[0102] Machine learning model 1.1: Supports CPU and OpenVino runtime;
[0103] Machine Learning Model 1.2: Supports GPU and TensorT runtime;
[0104] Machine Learning Model 1.3: Supports CPU, TensorFlow, or OpenVINO runtime.
[0105] S902, the base station downloads the machine learning model from the OMC.
[0106] In this embodiment of the invention, after receiving the model download instruction, the base station downloads a machine learning model that matches its supported runtime from the OMC based on HTTPS.
[0107] For example, assuming the base station only contains a CPU, when TensorFlow runs, the base station downloads machine learning model 1.0.
[0108] S903, base station deployment of machine learning models.
[0109] In this embodiment of the invention, after downloading the image package of the machine learning model, the base station deploys the machine learning model, i.e., activates the machine learning model, deserializes the machine learning model, and constructs the context for the inference and execution of the machine learning model. The base station can deploy the machine learning model in an independent microservice, or it can deploy the communication service and the machine learning model in the same microservice.
[0110] S904: Base stations provide inference services for communication services based on machine learning models.
[0111] Among them, base stations can provide REST-type or RPC-type inference services for communication services based on machine learning models.
[0112] When providing inference services, base stations can invoke at least two machine learning models deployed on the base station to collaborate and complete the inference service based on a multi-model collaboration strategy. This multi-model collaboration strategy can be received through the ServingSDK API. For example, see... Figure 10 As shown, a multi-model collaborative inference runtime is built based on the Serving SDK. The first model (model 1) is used for pre- and post-processing data handling and runs in TensorFlow and the CPU runtime. The second model (model 2) can only start inference after meeting specific conditions; it runs in TensorFlow and the CPU runtime. The third model (model 3) is executed n times in a loop and runs in OpenVINO and the CPU runtime. The fourth model (model 4) runs in a custom runtime framework (Custom) and the CPU runtime under specific conditions, used for post-processing data handling, and finally outputs the inference result.
[0113] The S905 base station polls the OMC's MODEL STORE to discover new versions of machine learning models.
[0114] The base station can periodically poll the OMC's MODEL STORE to find out if there are any new versions of the machine learning model.
[0115] S906: The base station decides for itself when to download, activate, and load the new version of the machine learning model.
[0116] After discovering a new version of the machine learning model, the base station decides on the timing of downloading, activating, and loading the new version of the machine learning model based on the business situation, thereby completing the switch between the old and new versions and maximizing the continuity of wireless services and minimizing impact.
[0117] In this embodiment of the invention, the base station can also revert the version of the machine learning model to a specified version when it receives a rollback instruction sent by the OMC.
[0118] In this embodiment of the invention, the steps on the base station side of the above-described communication service provision method can be executed by a microservice within the base station, and the machine learning model is deployed in this microservice. For example, see [link to documentation]. Figure 11 As shown, Figure 11 This is an architecture diagram of a Serving within a base station. It includes an HTTP server, which downloads machine learning models matching its supported runtimes from the OMC (Operating Management Center) using the HTTP server. The Serving manages these models using a ModelManager, which handles tasks such as downloading, activating, and loading them. The ModelManager can also poll the OMC's Model Store to check for newer versions of the machine learning models. The Serving can also provide RPC-type inference services for communication services based on these machine learning models through an RPC server. In this embodiment, multiple different runtimes can be pre-encapsulated within the Serving based on the base station's hardware architecture. This means the Serving can support multiple different runtimes. It should be noted that supporting different runtimes essentially means deploying different Serving images. For example... Figure 11 In Serving, the runtime environments for CPU / GPU and TensorFlow, GPU and TensorFlow, and CPU and OpenVINO are encapsulated. In this embodiment of the invention, the Serving runtime can be configured by the user, for example... Figure 11 In Serving, the device and custom runtime environments can be configured by the user. This means that users can customize the inference runtime environment based on the specific hardware architecture.
[0119] The communication service provision method provided in this embodiment of the invention allows a base station to provide inference services for communication services through a machine learning model deployed on its own. In some implementations, because the base station provides inference services for communication services through the machine learning model, the advantages of the machine learning model itself in data processing improve the efficiency of network operation. At the same time, the application of the machine learning model replaces manual processing in some aspects, saving costs. Furthermore, the base station can directly perform inference based on the machine learning model after receiving mobile phone data and complete the corresponding decisions. Compared with the scheme where the base station sends the collected data to a higher-level network (such as OMC) for decision-making, it can not only reduce latency but also improve data security.
[0120] Example 3
[0121] Based on Embodiments 1 and 2, this invention provides a communication service providing apparatus, including a service module, for providing communication services through a machine learning model.
[0122] In this embodiment of the invention, machine learning models are applied to communication applications, such as network planning and optimization. These models can provide at least one of the following services for communication services: site health check service, fault prediction service, base station planning service, radio frequency planning service, traffic prediction service, load prediction service, KPI (Key Performance Indicator) prediction service, coverage optimization service, parameter optimization service, and equipment energy saving service. In other words, in this embodiment, machine learning models can be applied to site health checks, fault prediction, base station planning, radio frequency planning, traffic prediction, load prediction, KPI prediction, coverage optimization, parameter optimization, and equipment energy saving. The machine learning models can provide REST or RPC-type inference services.
[0123] In this embodiment of the invention, the machine learning model can be a machine learning model designed based on at least one machine learning model running framework among Keras, TensorFlow, PyTorch, Caffe, and MxNet.
[0124] In this embodiment of the invention, before providing inference services for communication services through the machine learning model, the base station needs to download and deploy the machine learning model. When deploying the machine learning model, it can be deployed in a separate microservice. Alternatively, where computing resources are limited and requirements for latency and power consumption are more stringent, to maximize performance, the machine learning model and communication services can be deployed in the same microservice. In this way, the communication services can directly implement the model's inference requests through function calls.
[0125] In this embodiment of the invention, based on Embodiments 1 and 2, a machine learning model providing device is further provided, including a storage module for storing machine learning models for use by a communication service providing device. The machine learning model providing device may be an OMC (Optical Machine Control Unit).
[0126] The storage module stores trained machine learning models, which can be published in binary form. Communication service providers can download these models from the machine learning model provider. In other words, in this embodiment, the offline-trained machine learning model is uploaded to the machine learning model provider for download. It should be understood that, based on the needs of communication services, the network structure of the machine learning model can be designed using at least one of the following machine learning model runtime frameworks: Keras, TensorFlow, PyTorch, Caffe, and MxNet. The training dataset is then connected to complete the training of the machine learning model. The trained model can be easily exported to a compilation and optimization toolchain. After model optimization such as model pruning, layer fusion, and accuracy reduction, an optimized binary machine learning model is published and uploaded to the machine learning model provider. The machine learning model provider can store the machine learning model in a MODEL STORE.
[0127] It is understandable that different communication service providers have different hardware environments and support different runtime environments for machine learning models. The runtime environment for a machine learning model includes, but is not limited to, the hardware runtime environment and / or the machine learning model runtime framework. The hardware runtime environment can be hardware such as a GPU (Graphics Processing Unit) or a CPU (Central Processing Unit). For example, some communication service providers only include a CPU, so machine learning models that only support GPU runtimes cannot run on such devices. To ensure that different communication service providers can download machine learning models that match their supported runtime environments, in this embodiment of the invention, for the same machine learning model, at least two types of machine learning models can be published in the machine learning model provider. Different types of machine learning models support different runtime environments; that is, if two machine learning models have different runtime frameworks, their types are different; if two machine learning models support different hardware runtime environments, their types are different; if both the machine learning model runtime framework and the supported hardware runtime environment are different, their types are different. In this way, the communication service provider can download a machine learning model that matches its own hardware environment from at least two types of machine learning models. For example, the communication service provider can download a machine learning model that matches its own hardware environment from at least two types of machine learning models.
[0128] The process of a communication service provider downloading a machine learning model from a machine learning model provider may include: the machine learning model provider sending a model download instruction to the communication service provider; and the communication service provider downloading the machine learning model from the machine learning model provider after receiving the instruction. The communication service provider may download the machine learning model from the machine learning model provider using at least one of the following methods: SSH, HTTPS, FTPS, etc. After downloading the machine learning model, if the download is successful, the communication service provider may notify the machine learning model provider of the success; if the download fails, it will notify the machine learning model provider of the failure and attempt to resume the download process.
[0129] After downloading a machine learning model, the communication service provider needs to activate and load the model to complete its deployment. Activation refers to enabling the machine learning model, and loading refers to deserializing the model to achieve this. Furthermore, if a new version of the machine learning model becomes available after deployment, the communication service provider can upgrade the model. Alternatively, if a rollback to a specific version is needed after deployment, the communication service provider can revert to that version.
[0130] In this embodiment of the invention, the activation, loading, version upgrade, and version rollback of the machine learning model on the communication service providing device can be controlled by the machine learning model providing device. For example, the control process of the machine learning model on the communication service providing device by the machine learning model providing device includes: the machine learning model providing device sending a model control instruction to the communication service providing device; and the communication service providing device controlling the machine learning model according to the model control instruction after receiving the model control instruction. In this embodiment of the invention, the model control instruction includes, but is not limited to, at least one of the following control instructions: activation instruction, loading instruction, version upgrade instruction, and rollback instruction. The communication service providing device activates the machine learning model according to the activation instruction, loads the machine learning model according to the loading instruction, upgrades the version of the machine learning model according to the version upgrade instruction, and rolls back the version of the machine learning model to a specified version according to the rollback instruction.
[0131] In this embodiment of the invention, the communication service provider can also poll the machine learning models in the machine learning model provider to automatically discover new versions of the machine learning models. For example, when the machine learning model is stored in the MODELSTORE, the communication service provider can poll the MODELSTORE in the machine learning model provider to discover new versions of the machine learning models. After discovering a new version of the machine learning model, the communication service provider can decide when to download, activate, and load the new version of the machine learning model based on its own business situation, network status, etc.
[0132] In this embodiment of the invention, a Serving SDK (Software Development Kit) for machine learning models can be configured on the communication service providing device to manage the machine learning models deployed on the communication service providing device. The Serving SDK includes at least one of the following functions: model uploading, model upgrading, model scheduling, model inference, model monitoring, and runtime isolation.
[0133] In this embodiment of the invention, the Serving SDK includes an API for receiving control information for machine learning models and providing corresponding feedback based on the control information. In other words, users can control the machine learning models on the communication service provider device through the Serving SDK's API. The API can be a C-based API or a C++-based API. In this embodiment, the control information can be a query command, meaning users can use the API to query the machine learning models deployed on the communication service provider device, as well as the running status of each machine learning model. The control information can also be a customized solution for the machine learning model; that is, the API can be used to extend the customized solution for the machine learning model. The customized solution includes, but is not limited to, extending the functionality of the machine learning model, multi-model collaboration strategies, runtime, etc. In other words, in this embodiment, users can develop and customize machine learning models through the API.
[0134] In this embodiment of the invention, the Serving (microservice) in the communication service provider can manage the machine learning model by calling the Serving SDK.
[0135] It should be noted that, in this embodiment of the invention, the communication service providing device may deploy one machine learning model or at least two machine learning models. When at least two machine learning models are deployed on the communication service providing device, the machine learning models on the communication service providing device can independently complete the inference service, or the machine learning models on the communication service providing device can also cooperate with each other to complete the inference service. Specifically, the communication service providing device can call at least two machine learning models from the at least two machine learning models deployed on the communication service providing device to cooperate and provide inference services for the communication service according to a multi-model cooperation strategy.
[0136] In this embodiment of the invention, a multi-model collaboration strategy can be set within the machine learning model. That is, the multi-model collaboration strategy is set within the algorithm of the machine learning model. During the execution of the machine learning model, related machine learning models are automatically invoked for collaboration based on this strategy to complete the inference service. For example, in a TensorFlow-based machine learning model, a control operation (OP) is used to set the multi-model collaboration strategy within the algorithm. This can be achieved through operations like Switch, Merge, Enter, Leave, and Next, similar to if-else, while, and for statements in high-level languages. During training, each machine learning model can be trained independently. During deployment, the machine learning models, including the control operation, can be deserialized. During inference, the machine learning models, including the control operation, execute the control operation, thereby invoking other related machine learning models for collaborative inference.
[0137] Alternatively, the communication service provider can receive multi-model collaboration strategies through the API in the Serving SDK. This means users can control at least two machine learning models on the communication service provider to collaborate via the API interface. The multi-model collaboration strategy can be an imperative strategy, i.e., using native C++ to edit the multi-model collaboration strategy and control the collaboration of multiple machine learning models; or it can be a declarative strategy, i.e., using the DSL provided by the Serving SDK to construct a multi-model orchestrator, thereby invoking at least two machine learning models to collaborate. The most basic unit of DSL scheduling is a ScheduleAction. Each ScheduleAction can represent data preprocessing or inference of a sub-model. Collaboration methods include, but are not limited to: sequential execution, branch execution, and iterative execution.
[0138] The communication service provider and machine learning model provider provided in this embodiment of the invention provide inference services for communication services through machine learning models deployed on their own. In some implementations, because the communication service provider provides inference services for communication services through machine learning models, the advantages of machine learning models in data processing improve the efficiency of network operation. At the same time, the application of machine learning models replaces manual processing in some aspects, saving costs. Furthermore, after collecting data from mobile phones, the communication service provider can directly perform inference based on machine learning models to complete corresponding decisions. Compared with the scheme where the communication service provider sends the collected data to a higher-level network (such as the machine learning model provider) for decision-making, it can not only reduce latency but also improve data security.
[0139] Example 4:
[0140] This invention provides a base station, which includes a first processor 1201, a first memory 1202, and a first communication bus 1203;
[0141] The first communication bus 1203 is used to realize the connection and communication between the first processor 1201 and the first memory 1202;
[0142] The first processor 1201 is used to execute one or more programs stored in the first memory 1202 to implement the steps on the base station side in the communication service provision method of Embodiment 1 and Embodiment 2 described above.
[0143] This invention provides a server, which may be an OMC server, comprising: a second processor 1301, a second memory 1302, and a second communication bus 1303;
[0144] The second communication bus 1303 is used to realize the connection and communication between the second processor 1301 and the second memory 1302;
[0145] The second processor 1301 is used to execute one or more programs stored in the second memory 1302 to implement the steps on the OMC side of the communication service provision method in Embodiments 1 and 2 above.
[0146] This embodiment also provides a storage medium for storing one or more computer programs, which can be executed by a processor to implement at least one step on the base station side or OMC side in the communication service provision method of Embodiments 1 and 2 above.
[0147] The storage medium in this embodiment of the invention includes volatile or non-volatile, removable or non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, computer program modules or other data). Storage media include, but are not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), flash memory or other memory technologies, CD-ROM (CompactDisc Read-Only Memory), DVD or other optical disc storage, cartridges, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible by a computer.
[0148] The base station, server, and storage medium provided in this embodiment of the invention enable the base station to provide inference services for communication services through a machine learning model deployed on its own. In some implementations, because the base station provides inference services for communication services through the machine learning model, the advantages of the machine learning model itself in data processing improve the efficiency of network operation. At the same time, the application of the machine learning model replaces manual processing in some aspects, saving costs. Furthermore, the base station can directly perform inference based on the machine learning model after receiving mobile phone data and complete the corresponding decision. Compared with the scheme where the base station sends the collected data to the higher-level network (such as OMC) for decision-making, it can not only reduce latency but also improve data security.
[0149] Therefore, those skilled in the art should understand that all or some of the steps, systems, and devices disclosed above, as well as the functional modules / units, can be implemented as software (which can be implemented using computer program code executable by a computing device), firmware, hardware, and suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or as hardware, or as integrated circuits, such as application-specific integrated circuits (ASICs).
[0150] Furthermore, as is known to those skilled in the art, communication media typically contain computer-readable instructions, data structures, computer program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium. Therefore, this invention is not limited to any particular combination of hardware and software.
[0151] The above description, in conjunction with specific implementation methods, provides a further detailed explanation of the embodiments of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.
Claims
1. A method for providing communication services, comprising: The base station calls at least two machine learning models from its deployed machine learning models to collaborate in order to provide inference services for communication services, based on a multi-model collaboration strategy. Each deployed machine learning model is a machine learning model downloaded by the base station from the machine learning model providing device and matched with the runtime supported by the base station; the runtime includes the machine learning model's runtime framework and hardware runtime environment; The machine learning model providing device is used to publish at least two different runtime machine learning models for the same service function. Each machine learning model is used to call other relevant machine learning models for collaborative inference during inference, according to the control flow set for it by the multi-model collaboration strategy.
2. The communication service provision method as described in claim 1, characterized in that, The machine learning model is used to provide at least one of the following services for communication services: site health check service, fault prediction service, base station planning service, radio frequency planning service, traffic prediction service, load prediction service, key performance indicator (KPI) prediction service, coverage optimization service, parameter optimization service, and equipment energy saving service.
3. The communication service provision method as described in claim 1, characterized in that, The machine learning model includes a machine learning model designed based on at least one machine learning model running framework among Keras, TensorFlow, PyTorch, Caffe, and MxNet.
4. The communication service provision method as described in claim 1, characterized in that, The base station deploys the machine learning model in an independent microservice.
5. The communication service provision method as described in claim 1, characterized in that, The base station deploys the machine learning model and communication services in the same microservice.
6. The communication service provision method as described in claim 1, characterized in that, The base station extends the customized solution for the machine learning model through the Application Programming Interface (API) in the Serving SDK, a service software development kit for machine learning models.
7. The communication service provision method as described in claim 1, characterized in that, The multi-model collaboration strategy is set in the machine learning model.
8. The communication service provision method as described in claim 1, characterized in that, Also includes: The base station receives the multi-model cooperation strategy via API.
9. The communication service provision method as described in claim 8, characterized in that, The multi-model collaboration strategy includes an imperative multi-model collaboration strategy or a declarative multi-model collaboration strategy.
10. The method for providing communication services as described in any one of claims 1-9, characterized in that, The method further includes: The base station downloads the machine learning model and deploys it on itself.
11. The communication service provision method as described in claim 10, characterized in that, The machine learning model is provided by the Operation and Maintenance Center (OMC). The base station download machine learning model includes: The base station downloads the machine learning model from the OMC.
12. The communication service provision method as described in claim 11, characterized in that, For the same machine learning model, the OMC includes at least two types of machine learning models, and the different types of machine learning models support different runtime environments.
13. The communication service provision method as described in claim 11, characterized in that, The base station downloads the machine learning model from the OMC, including: After receiving the model download instruction sent by the OMC, the base station downloads the machine learning model from the OMC.
14. The communication service provision method as described in claim 11, characterized in that, Also includes: The OMC sends model control commands to the base station; After receiving the model control command, the base station controls the machine learning model according to the model control command.
15. The method for providing communication services as described in claim 14, characterized in that, The model control commands include at least one of the following: activation command, loading command, version upgrade command, and rollback command; The control of the machine learning model according to the model control instructions includes at least one of the following methods: The machine learning model is activated according to the activation instruction; The machine learning model is loaded according to the loading instruction; Upgrade the version of the machine learning model according to the version upgrade instruction; The machine learning model version is reverted to the specified version according to the rollback instruction.
16. The communication service provision method as described in claim 11, characterized in that, The OMC includes a model store, which is used to store machine learning models; The base station discovers new versions of the machine learning model by polling the MODEL STORE.
17. The method for providing communication services as described in claim 16, characterized in that, After discovering a new version of the machine learning model, the base station also includes: The base station decides for itself when to download, activate, and load the new version of the machine learning model.
18. A communication service providing apparatus, the communication service providing apparatus comprising: A service module is used to enable a base station to call at least two machine learning models from multiple machine learning models deployed by itself for collaboration through a multi-model collaboration strategy, in order to provide inference services for communication services; each deployed machine learning model is a machine learning model downloaded by the base station from a machine learning model providing device that matches the runtime supported by the base station; the runtime includes the machine learning model's runtime framework and hardware runtime environment; wherein, the machine learning model providing device is used to publish at least two different machine learning models of the runtime for machine learning models with the same service function; wherein, each machine learning model is used to call other relevant machine learning models for collaborative inference according to the control flow set for it by the multi-model collaboration strategy during inference.
19. A machine learning model providing apparatus, the machine learning model providing apparatus comprising: A storage module is used to store machine learning models, allowing a base station to call at least two machine learning models from its deployed multiple machine learning models for collaboration according to a multi-model collaboration strategy, in order to provide inference services for communication services. Each machine learning model is a machine learning model downloaded by the base station from a machine learning model providing device that matches the runtime supported by the base station. The runtime includes the machine learning model's runtime framework and hardware runtime environment. The machine learning model providing device is used to publish at least two different machine learning models with the same service function for machine learning models using the same runtime. Each machine learning model is used to call other relevant machine learning models for collaborative inference according to the control flow set for it by the multi-model collaboration strategy during inference.
20. A base station, the base station comprising: A first processor, a first memory, and a first communication bus; The first communication bus is used to realize the connection and communication between the first processor and the first memory; The first processor is configured to execute one or more programs stored in the first memory to implement the steps on the base station side of the communication service provision method as described in any one of claims 1 to 17.
21. A server, the server comprising: A second processor, a second memory, and a second communication bus; The second communication bus is used to realize the connection and communication between the second processor and the second memory; The second processor is used to execute one or more programs stored in the second memory to implement the steps on the OMC side of the communication service provision method as described in any one of claims 12-17.
22. A storage medium storing one or more programs, the one or more programs being executable by one or more processors to implement the steps on the base station side of the communication service provision method as claimed in any one of claims 1 to 17, or the steps on the OMC side of the communication service provision method as claimed in any one of claims 12 to 17.