GJR2391500R1210 81EU01 ABB通用输入模块

　　LCM 的一个重要方面是它代表了供应商和 CSP 之间的责任、问责和所有权结构。这种结构是供应商和 CSP 之间业务模型的基线，准确地构建了供应商在 LCM 流程中交付的内容。

　　图 4 中的浅橙色和绿色背景色分别突出了 CSP 和供应商的职责。软件或软件/硬件实体在适应/接受步骤中与支持合同一起交付，在某些情况下，还包括用于集成和部署的服务。

　　在 RAN 自动化解决方案中使用 AI/ML 模型需要在 LCM 流程中引入模型训练步骤。如何将模型训练添加到 LCM 有四种主要替代方法，每种方法都对供应商和 CSP 之间的责任划分产生影响。

　　种选择是供应商在适应/接受步骤中以软件实体的形式提供一个全局模型（即所有 CSP 的相同模型）。对于某些用例，全局模型仍然可以考虑本地上下文，并且在创建可以适应不同部署的高度灵活的自动化功能方面非常强大。在这种情况下，所有培训都是供应商的责任，并且发生在培训步骤中。

　　第二种选择是让供应商在适应/接受步骤中以针对不同用途（例如，CSP 特定或地理特定）定制的软件实体的形式提供本地模型。本地培训是供应商的责任，发生在培训步骤中。这种完整的模型训练替代方案需要访问本地数据，重要的是要意识到维护不同软件版本的成本可能会变得很大。因此，这种替代方案最适合在每个 CSP 的几个地方集中推理的场景，其中只有一个或几个不需要频繁重新训练的 ML 模型。在每个 CSP 需要在数千个地方进行分布式推理的场景中，需要每隔一周重新训练一次（例如），这种模型训练不是选择。

　　第三种选择是让供应商提供一个可以在其他数据集上重新训练的全局模型。在适应/接受步骤中，供应商以软件实体的形式提供模型以及有关如何重新训练和它的信息。CSP 负责将模型重新训练为一组本地模型，这将适应/接受步骤扩展为包括训练。在这些情况下，尚不清楚供应商可以为现场性能和支持承担多少责任。因此，在责任得到解决之前，不建议将其作为商业部署的方向。

　　第四种选择是供应商以软件的形式提供经过基础训练的模型，该软件旨在在部署后根据本地数据自动重新训练。我们将此称为嵌入式培训，培训对 CSP 是透明的。在这种情况下，培训是供应商的责任，并在培训步骤和部署的软件中自主进行。这是一条通往完全自治系统的道路，同时保持供应商和 CSP 之间当前的业务关系完好无损。

　　云 RAN 实施将对 LCM 流程施加超出 AI/ML 引入的额外更改。基于网络中的本地和时间变化（例如负载），云原生、基于微服务的架构将能够以微服务的形式非常动态地部署和实例化功能。在负载移动的网络中，随着负载的移动，此功能还应扩展到在网络的不同部分实例化/扩展微服务。由于变化的动态性，这些过程需要自动化，这意味着手动部署步骤的一部分是自动化的，并由供应商提供的功能控制。

　　随着虚拟化和编排趋势的发展，几乎所有的部署、扩展、测试和实例化都可能自动且高度动态地发生。届时，CSP 的职责将从手动部署软件转移到监控 RAN 自动化解决方案实现 RAN 意图的程度。

　　以下是我司【主营产品】，有需要可以发来帮您对比下价格哦!

　　主营：世界品牌的plc 、dcs 系统备件 模块

　　①allen-bradley(美国ab)系列产品》

　　②schneider(施耐德电气)系列产品》

　　③general electric(通用电气)系列产品》

　　④westinghouse(美国西屋)系列产品》

　　⑤siemens(西门子系列产品)》

　　⑥销售abb robots. fanuc robots、yaskawa robots、kuka robots、mitsubishi robots、otc robots、panasonic robots、motoman robots。

　　⑦estinghouse(西屋)： ovation系统、wdpf系统、max1000系统备件。

　　⑧invensys foxboro(福克斯波罗)：i/a series系统，fbm(现场输入/输出模块)顺序控制、梯形逻辑控制、事故追忆处理、数模转换、输入/输出信号处理、数据通信及处理等。invensys triconex: 冗余容错控制系统、基于三重模件冗余(tmr)结构的现代化的容错控制器。

　　⑨siemens(西门子)：siemens moore， siemens simatic c1，siemens数控系统等。

　　⑩bosch rexroth(博世力士乐)：indramat，i/o模块,plc控制器,驱动模块等。

　　◆motorola(摩托)：mvme 162、mvme 167、mvme1772、mvme177等系列。

　　plc模块，可编程控制器，cpu模块，io模块，do模块，ai模块，di模块，网通信模块，

　　以太网模块，运动控制模块，模拟量输入模块，模拟量输出模块，数字输入模块，数字输出

　　模块，冗余模块，电源模块，继电器输出模块，继电器输入模块，处理器模块。

　　我们的优势是：全新原装，，供给一年质保!本公司所有产品都经过严格检测，欢迎询价，收购。只需您有诚心，本公司将会给你供给一个比同行优势的价格，共同拿下单子。

　　An important aspect of the LCM is that it represents a structure of responsibility, accountability and ownership among vendors and CSPs. This structure is the baseline for the business model between vendors and CSPs, structuring exactly what is delivered by the vendor in the LCM process.

　　The light orange and green background colors in Figure 4 highlight the responsibilities of the CSP and the vendor respectively. Software or software/hardware entities are delivered in the adapt/accept step together with support contracts and, in some cases, professional services for integration and deployment.

　　Using AI/ML models in the RAN automation solution requires the introduction of a model training step to the LCM process. There are four main alternatives for how to add model training to the LCM, each with implications on the responsibility split between the vendor and the CSP.

　　The first alternative is for the vendor to deliver a global model (that is, the same model for all CSPs) in the form of software entities in the adapt/accept step. A global model can, for some use cases, still allow for consideration of local context and can be very powerful in creating highly flexible automation functionality that can adapt to different deployments. In this case, all training is the responsibility of the vendor and occurs in the train step.

　　The second alternative is for the vendor to deliver local models in the form of software entities tailored for different uses (CSP-specific or geo-specific, for example) in the adapt/accept step. Local training is the responsibility of the vendor and occurs in the train step. This full model training alternative requires access to local data, and it is important to be aware that the cost of maintaining different software versions could become substantial. As a result, this alternative is most appropriate for scenarios with centralized inference in a few places per CSP where there is only one or just a few ML models that do not require frequent retraining. In scenarios with distributed inference in thousands of places per CSP that require retraining every other week (for example), this model training would not be the best alternative.

　　The third alternative is for the vendor to deliver a global model that can be retrained on additional data sets. In the adapt/accept step, the vendor delivers the model in the form of software entities together with information about how to retrain and evaluate it. The CSP is responsible for retraining the model to become a set of local models, which expands the adapt/accept step to include training. In these scenarios, it is unclear how much responsibility the vendor can take for in-field performance and support. Therefore this is not recommended as a direction commercial deployment until responsibilities have been resolved.

　　The fourth alternative is for the vendor to deliver a base-trained model in the form of software that is designed to be automatically retrained on local data after deployment. We refer to this as embedded training, and the training is transparent to the CSP. In this case, the training is the responsibility of the vendor and occurs both in the train step and autonomously in the deployed software. This is a path toward a fully autonomous system, while keeping the current business relation between vendor and CSP intact.

　　A cloud RAN implementation will impose additional changes to the LCM process that go beyond those introduced by AI/ML. A cloud-native, microservice-based architecture will enable the possibility to very dynamically deploy and instantiate functionality in the form of microservices, based on local and temporal changes in the network, such as load. In a network with moving load, this capability should also extend to instantiating/scaling microservices in different parts of the network as load moves around. Because of the dynamics of the changes, these processes need to be automated, meaning that parts of the manual deployment step are automated and governed by functionality provided by the vendor.

　　As the trend of virtualization and orchestration evolves, it is probable that nearly all deployment, scaling, canary testing and instantiation will happen automatically and highly dynamically. At that point, the CSPs’ responsibility will move from the manual deployment of software to monitoring how well the RAN automation solution fulfills the RAN intents.