AI Enabled Control of Distributed Generation

点击翻译按钮获取中文摘要

The control of power electronic interfaces (inverters) for renewable energy and storage must be designed to operate stably in a power system with widely varying operating conditions. This is becoming more challenging as inertia and conventional generation are being increasingly replaced with inverter-based generation, and interoperability of different inverter control systems is not guaranteed. This project will explore how artificial intelligence (AI) can be used to tune and adapt inverter control on-line by estimating the state of the local grid using measurements made by an inverter as part of its normal control. These estimates are especially important during abnormal operating conditions eg faults, where cascade failures can occur with inverter based generation. AI will be used to continually monitor local grid conditions and determine changes to control functions and modes especially when system faults occur. Two approaches will be employed. The first will use AI for grid status estimation, e.g., system-wide impedance based on partial information from local measurements. Data-driven AI methods, e.g., deep learning methods, will be exploited as a replacement for its deterministic counterparts for responding to inverter and power system controls, especially for the non-linear operating regions where conventional techniques fail. Experimental data will be collected from typical inverter systems available in the FlexElec laboratory and used to train AI models.The second approach will exploit model-driven AI methods, where the state-of-the-art phase-amplitude mathematical framework will be exploited as the model knowledge for the design and implementation of the AI architecture. The phase-amplitude mathematical framework will be further developed through collaboration with Maths Department, and the framework will be adapted and tailored to aid the design, training and implementation of the AI model. By incorporating knowledge of the power system structure into the AI model, the proposed approach will improve the grid status estimation for locally connected power converters (and inherently enhance their operation). Both approaches will be evaluated experimentally in the FlexElec laboratory and compared to conventional grid following and grid forming technologies.