FPGA-based Real-time Simulation of Grid-Connected Distributed Solar Systems using Multi-Agent Control

基于 FPGA 的多智能体控制并网分布式太阳能系统实时仿真

• 批准号: RGPIN-2022-03004
• 负责人: Chandra, Ambrish
• 金额: $ 4.01万
• 依托单位: École de technologie supérieure
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2022
• 资助国家: 加拿大
• 起止时间: 2022-01-01 至 2023-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=750689
• 关键词: FPGA; based; Real; time; Simulation

High penetration of renewable energy sources in distribution power systems is causing challenges to the voltage regulation of grids. Voltage instability may trigger serious damages in power networks. Current trend in control of power systems is moving toward the use of distributed, flexible, and robust structures. Applying an intelligent agent-based control system is an effective way for solving complex problems in parallel. The Alternating Direction Method of Multipliers (ADMM) can be used as a theoretical framework to design distributed computational systems. ADMM-based algorithm is an iterative optimization method able to solve complex problems by decomposing them into multiple sub-problems in a fully distributed way. Real-time simulation is an efficient way that allows researchers to test complex designed controllers in a secure environment. However, the computational time in the real-time (RT) simulators increase with the complexity of the models. In recent years, despite the identified advantages of the application of multi-agent control technique in power systems, only a few works have been done to modify the ADMM algorithm to fully utilize the parallelized hardware structure of the Field Programmable Gate Arrays (FPGAs) in real-time. FPGA-based platforms are used for Hardware-in-the-Loop (HIL) testing as a solution for running more realistic models in less time. FPGAs offer high-performance simulations in parallel with less time step and higher flexibility. This program is an innovative application in real-time simulation of grid-connected PV systems integrated to an optimal voltage control based on multi-agent systems. Coordination between different agents will be implemented using FPGA. The main objective of the program is to design an innovative, simple, and robust application of FPGA in real-time control of PV systems connected to Hydro- Québec's networks to ensure optimal management of local energy as well as the safe injection of clean energy. The short-term objectives of the program are to: 1.Design and simulate an intelligent agent-based control algorithm based on ADMM. 2.Analyze the operation of proposed technique under high penetrations of PVs. 3.Develop the distributed optimization-based voltage control algorithm to accelerate the convergence of the algorithm. 4.Validate the method by implement it on real networks in large-scale. 5.Design a parallel simulation approach using real-time FPGAs ensuring minimum latency in the whole system division. The long-term objectives of the project are to: 1. Implement a low-cost and effective solution for voltage stability of the system under uncertainties during distribution and transmission level.  2.Develop an intelligent method for training the agents using neural networks aims to reduce the running time of distributed ADMM algorithms. 3.Develop a unique expertise in the management and control of local micro-networks.

可再生能源在配电系统中的高渗透率给电网的电压调节带来了挑战，电压不稳定可能会引发电网的严重损坏。当前电力系统控制的趋势正在朝着使用分布式、灵活和鲁棒结构的方向发展。应用基于智能代理的控制系统是并行解决复杂问题的有效方法。交替方向乘子法（ADMM）可以作为设计分布式计算系统的理论框架，是一种迭代优化方法。能够解决通过以完全分布式的方式将复杂问题分解为多个子问题来解决复杂问题是一种有效的方法，它允许研究人员在安全的环境中测试复杂设计的控制器，但是实时（RT）的计算时间。近年来，尽管多智能体控制技术在电力系统中的应用具有明显的优势，但对ADMM算法进行修改以充分利用电力系统的并行化硬件结构的工作却很少。现场可编程门基于 FPGA 的平台用于硬件在环 (HIL) 测试，作为在更短的时间内运行更真实模型的解决方案，FPGA 可以在更短的时间内提供高性能的并行仿真。该程序是基于多代理系统的最优电压控制的并网光伏系统实时仿真的创新应用，其主要目标是使用FPGA来实现。程序的目标是设计一种创新、简单且稳健的 FPGA 应用，实时控制连接到魁北克水电公司网络的光伏系统，以确保本地能源的优化管理以及清洁能源的安全注入。该计划的目的是： 1.设计并模拟基于 ADMM 的智能代理控制算法 2.分析所提出的技术在光伏发电高渗透率下的运行情况 3.开发基于分布式优化的电压控制算法以加速。 4. 通过在真实网络上大规模实施来验证该方法。 5. 使用实时 FPGA 设计并行仿真方法，确保整个系统划分的延迟最小。目标是： 1. 在配电和输电层面的不确定性下实现系统电压稳定性的低成本且有效的解决方案。 2. 开发一种使用神经网络训练代理的智能方法，旨在减少分布式 ADMM 算法的运行时间。 。 3.发展本地微网络管理和控制方面的独特专业知识。