通过LLM和Amazon Bedrock试剂集成增强诊断流程

Noodoe是电动汽车充电创新的全球领导者，提供先进的解决方案，使运营商能够优化其充电站运营，并提供卓越的用户体验。他们的通用充电站与所有电动汽车品牌兼容，并具有直观的支付选项，包括信用卡和Apple Pay。该公司采用Noodoe EV OS云管理系统，提供全天候的自动监控、诊断和维护，实现了99.83%的市场领先正常运行时间。Noodoe在超过15个国家开展业务，并坚定地致力于可持续发展，通过尖端技术和用户至上的方法改变电动汽车充电行业。

尽管拥有技术优势，Noodoe在帮助电站运营商优化性能和在不同市场中选择最具成本效益的电价策略方面遇到了关键挑战。传统系统缺乏有效处理大量实时和历史数据或提供个性化车站级建议的能力。这限制了运营商做出及时、明智决策的能力，导致电力成本上升、资产利用不足和客户体验不佳。这些低效不仅降低了盈利能力，还阻碍了在竞争激烈、快速发展的电动汽车充电环境中扩大规模的能力。

为了解决这个问题，Noodoe通过亚马逊基岩和亚马逊基岩代理集成了大型语言模型（LLM），以提供智能自动化、实时数据访问和多语言支持。这些人工智能驱动的工具分析使用模式、车站诊断和天气或电网状况等外部变量，以生成高度定制的定价建议。通过使用Amazon Bedrock的结构化编排和基于提示的推理，Noodoe为运营商提供了可操作的见解，以提高利润率，提高车站利用率，并允许他们为用户提供更具竞争力的价格，最终提高客户满意度。这项服务是通过订阅模式提供的，为Noodoe创造了一个新的、可扩展的收入来源，同时加强了其在电动汽车基础设施领域的领导地位和创新能力。

在这篇文章中，我们将探讨Noodoe如何使用人工智能和亚马逊基岩来优化电动汽车充电操作。通过集成LLM，Noodoe增强了车站诊断，实现了动态定价，并提供了多语言支持。这些创新减少了停机时间，最大限度地提高了效率，并提高了可持续性。继续阅读，了解人工智能如何改变电动汽车充电管理。

解决方案概述

Noodoe AI增强的诊断流程建立在一个多步骤的过程之上，该过程结合了数据收集、AI驱动的分析和无缝翻译，以实现全球可访问性，如下图所示。

The physical charging station network currently operates over 1,000 sites across more than 20 countries, with plans to expand by more than 50 additional sites by the end of 2025. As illustrated in the following image, it uses the EV Cloud and LLMs to generate relevant recommendations following backend processing.

The following screenshot shows an example of the results in the UI.

Overview of Noodoe AI-enhanced diagnostics

The following diagram illustrates the solution data flow.

To meet the feature requirements, the system operation process includes the following steps:

1. Charging data is processed through the EV service before entering the database.
2. The charging history data and pricing data are stored in the EV database.
3. Amazon EventBridge Scheduler periodically triggers the EV service to perform analysis.
4. The EV service calls the AI service to analyze historical data and provide pricing recommendations.
5. The AI service collects the organized historical data to prepare the prompt template.
6. This information, combined with appropriate prompts, is used in conjunction with Amazon Bedrock Agents as an AI-pricing agent to extract relevant information. The AI-pricing agent analyzes this combined data to identify daily peak and off-peak periods and provide recommendations for user pricing plans.
7. Optionally, if translation is needed for non-English users, these results from the AI-pricing agent are further processed through another Amazon Bedrock agent for translation.
8. Optionally, the translation agent uses Anthropic’s Claude Sonnet 3.5 on Amazon Bedrock to get the result in the corresponding language.
9. Finally, the AI service collects the results in the user’s language for formatting and other processing, then inserts them into a template to create a comprehensive report that is pushed to the user’s end.

In the following section, we dive deep into these steps and the AWS services used.

Architecture of Noodoe AI-enhanced diagnostics

Noodoe faced key challenges in building a globally scalable, reliable, and cost-efficient architecture. They needed a solution that could support rapid expansion, handle high data volumes, and deliver consistent performance across AWS Regions. Addressing these requirements required careful architectural planning to provide flexibility and resilience.

The following diagram illustrates the solution architecture Noodoe built to overcome these challenges to support global growth.

The EV charging optimization platform structures the data flow across multiple AWS services, providing efficient data ingestion, processing, and AI-driven decision-making. Amazon Elastic Kubernetes Service (Amazon EKS) retrieves data from Amazon DocumentDB, processes it, and invokes Amazon Bedrock Agents for reasoning and analysis. This structured data pipeline enables optimized pricing strategies and multilingual customer interactions. By using containerized applications, event-driven workflows, and AI capabilities, the system provides scalable and flexible insights to EV station operators.

Data ingestion and processing

EV charging stations send real-time charging data to AWS IoT Core, which acts as the initial entry point for data processing. The data is then transmitted to Amazon Managed Streaming for Apache Kafka (Amazon MSK) to facilitate high-throughput, reliable streaming. From Amazon MSK, data flows into Amazon EKS, where the EV service processes it before storing the charging history and trend records in DocumentDB. This structured storage provides efficient retrieval for analysis and prediction.

AI-powered pricing analysis

To optimize pricing strategies, Amazon EventBridge triggers a pricing prediction function at regular intervals. This function retrieves historical charging data from DocumentDB and sends it, along with predefined prompts, to the Amazon Bedrock AI-pricing agent. The AI agent, powered by Anthropic’s Claude on Amazon Bedrock, evaluates station usage trends, peak and off-peak periods, and pricing inefficiencies to generate optimal pricing recommendations. Although the pricing agent doesn’t access an Amazon Bedrock knowledge base or trigger action groups, it uses preprocessing and post processing features to refine predictions and improve decision-making.

Multilingual support and report generation

If translation is required, the pricing analysis results are forwarded to the Amazon Bedrock translate agent, which converts the insights into the operator’s preferred language. The translated and structured data is then formatted into a predefined report template and stored in a designated database for later retrieval. This provides seamless access to actionable insights across diverse markets.

UI, monitoring, and performance optimization

Operators access the system through a web-based UI, with Amazon Route 53 and Amazon CloudFront providing fast and efficient content delivery. An Application Load Balancer distributes incoming requests across multiple EKS instances, providing high availability. To optimize performance, Amazon ElastiCache accelerates data retrieval while reducing database load. For system monitoring and observability, Amazon CloudWatch provides additional monitoring and observability. The administrator of Noodoe uses Amazon Managed Service for Prometheus and Amazon Managed Grafana for system monitoring and visualization.

This architecture empowers Noodoe with an AI-driven, scalable, and intelligent EV charging management solution, enhancing station utilization, revenue optimization, and customer experience worldwide.

总结

Noodoe AI增强的诊断流程通过整合Amazon Bedrock Agents、基于规则的自动化、实时用户输入和LLM驱动的洞察力来改变电动汽车充电操作，从而做出更明智的决策。在全面的知识库和简化的API的支持下，该解决方案使运营商能够自动化工作流程，优化定价，并大规模提高车站性能。知识库的持续扩展、工作流程的优化和现实世界的测试进一步提高了效率和可靠性。这种方法使收入增加了15%，实施时间缩短了10%。持续的反馈和清晰的文档使用户能够有效地使用人工智能驱动的诊断，实现更智能的充电管理。

Noodoe产品副总裁Roman Kleinerman表示：“随着客户使用我们的Al解决方案来优化定价策略，我们的收入增长了10-25%，具体取决于站点的位置和数量。”

Noodoe致力于提供更智能、更智能的电动汽车充电服务，使最终用户和运营商都受益。目前，Noodoe在20多个国家运营着1000多个充电站，并计划到2025年底再扩建50多个充电点。展望未来，该系统正在得到增强，通过结合需求、电网条件、时间和天气等因素，支持近乎实时的动态定价优化。Amazon Bedrock Agents有助于实现这些智能功能，为动态定价、负载平衡和电网感知路由提供支持，以优化能源分配并引导用户前往最高效的站点。未来的增强功能将根据用户偏好提供个性化的收费建议和激励措施，为客户和运营商实现价值最大化。与亚马逊Bedrock合作，开始构建智能、人工智能驱动的电动汽车充电解决方案。