存算一体技术研究进展及其在电网中的应用探索

Development of Compute-in-memory Technology and Its Potential Implementation in Power Grid

基于数据驱动的人工智能方法在电力系统运维实际应用中展示出相当的优势,但目前人工智能技术在电力系统中还未能实现泛在落地应用,其关键原因之一为电力智能计算设备对人工智能模型的支撑能力不足。一方面,目前电力边、端侧设备资源受限,普遍存在计算能力不足等问题,无法支撑复杂电力人工智能模型的部署和运行;另一方面,由于电力系统规模的扩大和复杂程度的增加,电力云计算中心需要处理PB级海量数据并进行大规模电力调度计算,开始出现“算不动”的问题,难以满足电力系统快速响应需求,且耗能也不断攀升。存算一体技术是一种直接利用存储器进行数据处理的新型计算技术,可实现高算力、低功耗电力数据高效能处理,为解决新型电力系统难题提供了新的思路。基于此,详细归纳了存算一体技术的主流研究方向,阐释了存算一体技术在电网应用的可行性,提出了一些潜在电力应用场景,分析了实际应用中可能会面临的挑战,旨在为存算一体技术在电网的应用明确重点和方向。

The methods of data-driven artificial intelligence have shown considerable superiority in the practical application of power system operation and maintenance.However,at present,the artificial intelligence technology has not been widely applied in power systems.One of the key reasons is the insufficient ability of power computing equipment for supporting the AI(artificial intelligence)models.On one hand,the limited resources on the power edge and end devices result in the insufficient computing power commonly existing,which makes the deployment and operation of the complex power AI models unpractical.On the other hand,the expansion and increase of the complexity of the power systems,the power cloud computing centers have to process the PB level of mass data and carry out a large scale power dispatching calculations.The computing equipment appears“not being able to calculate”,which makes it difficult to meet the rapid response of the power system and the rising energy consumption.Compute-in-memory technology,a new computing paradigm that directly uses the memory for data processing,can realize the high computing power with low power consumption,providing a new path for solving the problems raised in new power systems.This article summarizes the mainstream researches of compute-in-memory technology in detail,and explains the feasibility of the application of compute-in-memory technology in the power grids.It also puts forward some potential power application scenarios,and analyzes the challenges that may be faced with in the actual application.The aim of this article is to clarify the focus and direction of the application of compute-in-memory technology in the power grids.