报告时间：2022年5月20日（星期五）下午14:50-17：30

报告形式：腾讯会议ID：557-662-205

（一）

报告时间：14:50-15:10

报告人：帅 航

报告题目：基于强化学习的电力系统优化运行与恢复

主要内容：

The climate change and uncertainty of renewable energy bring significant challenges to the economic and reliable operation of power grids. Conventional online optimization approaches for power systems require a forecast model. However, accurately forecasting renewable power generations is still a tough task. To achieve online scheduling of power systems that do not need a forecast model to predict the future PV/wind and load power sequences, we investigate the application of reinforcement learning (RL) approaches to tackle this challenge. Firstly, based on the recent development of approximate/adaptive dynamic programming (ADP) techniques, we proposed several ADP based microgrid and bulk energy system online optimization methods. Secondly, to further improve the online optimization performance, we proposed the MuZero based residential microgrid optimization algorithm, which significantly reduced the operational cost of the system. Besides, to solve the economic dispatch of microgrids with multiple energy storage systems, we proposed the Branching Dueling Q-Network based online optimization approach. Finally, regarding distribution grids restoration after extreme weather events, the AlphaZero based restoration strategy was developed to repair the damaged system as fast as possible.

报告人简介：帅航，博士，2013年毕业于武汉工程大学自动控制系，2015年，2019年分别获得华中科技大学电气信息检测技术专业硕士和电力系统及其自动化博士学位。主要研究方向为基于（深度）强化学习的电力系统优化运行与控制技术。针对可再生能源接入及气候变化给电力系统安全经济运行带来的诸多挑战，先后提出了多种基于（深度）强化学习的电力系统在线优化策略及故障恢复算法。作为科研骨干先后参与了国家重点研发计划项目、国家自然科学基金-海外及港澳学者合作研究基金项目、美国国家自然科学基金项目、美国能源部（DOE）项目等。在IEEE Transactions等国际电力和能源领域的顶级期刊上接收发表论文8篇，其中ESI高被引论文2篇（均为一作）；另在IEEE PES General Meeting等国际会议上发表论文5篇，并获国际会议最佳论文奖1次。

（二）

报告时间：15:10-15:30

报告人：刘 懿

报告题目：面向可靠性的电力电子变流器设计、控制与优化

主要内容：

State-of-the-art reliability studies are failure rate driven which are based on the probability models and lacked of the failure mechanism analysis. Therefore, the physical of failure (PoF) reliability analysis has become popular in recent years. Based on the bathtub curve, the reliability related research can be divided into infant failure, random failure and aging. For infant reliability issues, the early failure mechanism has been analyzed, and a weak stress accelerated aging test (WSAAT) method for the key components of power electronics converter has been proposed, which realized the pre-sorting of early failure of components with stress-strain curve. For the random failure, an non-invasive on-line monitoring method of converter components based on component characteristics are proposed. Also, a single-component on-line monitoring circuit has been designed, which overcomes the parasitic parameter issues, electromagnetic compatibility and insulation caused by the external monitoring circuit. For aging, the reliability-oriented design methodology has been illustrated, and the design of the converter from a reliability perspective, besides other considerations, has been performed.

报告人简介：刘懿，博士后，2011年毕业于西南交通大学电气工程及其自动化专业，2019年获得电力电子博士学位。主要致力于电力电子变流装置可靠性及运行能力提升方面的研究。在研究工作中，针对变流器关键元件早期失效，提出了电力电子关键元件弱应力加速老化试验方法，实现了元件早期失效的预分拣；针对变流装置运行过程中的可靠性问题，提出了基于元件特性的变流器元件在线监测方法以及单元件在线监测电路，克服了引入外部监测电路带来的寄生参数、电磁兼容与绝缘影响；针对装置长期运行可靠性，提出了面向可靠性的电力电子变流装置设计方法，极大地节省了变流装置的设计时间与成本。所研成果已发表多篇电力电子顶级期刊。

（三）

报告时间：15:30-15:50

报告人：陈启明

报告题目：多元时频分析理论与工业过程振荡检测及诊断研究

主要内容：

Many common data, such as those sampled from nature, human-beings, and industrial systems, contain valuable information. This report introduces a novel Multivariate Nonlinear Chirp Mode Decomposition (MNCMD). In contrast to most existing multivariate time-frequency decomposition approaches, the proposed MNCMD is capable of handling time-varying signal efficiently in an elegant variational optimization framework. The multivariate nonlinear chirp mode is defined based on the presence of a joint or common instantaneous frequency component among all channels of input signal. Then the objective function of MNCMD is defined as the sum of mode bandwidths across all signal channels. MNCMD can extract an optimal set of multivariate modes and their corresponding instantaneous frequencies without requiring more user-defined parameters than the original NCMD. The effectiveness and advantages of the proposed MNCMD are demonstrated by studying its mode-alignment, filter bank structure, quasi-orthogonality, the influence of channel number, noise robustness, and convergence. Specifically, we highlight the utility and superiority of the proposed method in different real-world applications, including the EEG signal processing and plant-wide oscillations causality analysis.

报告人简介：陈启明，博士，2017年毕业于华北电力大学自动控制系，2017年至今就读于浙江大学控制科学与工程专业。主要致力于现代信号处理与工业控制系统性能评估研究。在研究工作中，针对现有过程振荡监测方案零散化和对复杂信号分析能力不足问题，提出了基于快速调频模态分解算法的过程振荡检测-诊断-溯源一体化解决方案；针对复杂系统厂级振荡根源分析不清晰不准确问题，提出了多元信号分解与因果推理融合定位方法，能有效定位厂级振荡源；针对变分模态分解算法的参数依赖问题，提出了自整定变分模态分解算法，能提高信号分解算法自适应性和鲁棒性；针对现有信号分解与时频分析算法对时变多元信号处理能力有限的问题，提出了多元非线性调频模态分解算法，是首个能处理时变信号的多元信号分解算法；针对多元非线性调频模态分解算法复杂度较高的问题，提出了多元本征调频模态分解算法，将算法复杂度从O(N^3)将低至O(N)。部分成果已发表在控制工程和信号处理领域顶级期刊上。

（四）

报告时间：15:50-16:10

报告人：陈冠宇

报告题目：集成光电子器件及其在光互连中的应用

主要内容：

Among the big data era, the explosive growth of the data makes great challenge to the communication system. The optical fiber based communication technology is widely used in the long-distance communication networks to realize ultra-large capacity and ultra-high speed transmission. However, indifferent network notes, integrated and compact optical module now is the trend. On the other hand, the traditional electrical interconnect shows its disadvantages in big data center and high performance computer due to its huge power consumption and low capacity. On the contrast, the optical interconnect, with low latency, low power and large capacity is the best alternative choice. Integrated photonics devices are greatly preferred. During the past decade, silicon photonics (SiPs) based devices have experienced great development. Some newly developed integration platform, such as lithium niobate (LN) also attracts great attentions. In this presentation, some high performance photonics devices based on SiPs and LN will be introduced.

报告人简介：陈冠宇，博士，2013年毕业于中国地质大学，2018年获得华中科技大学博士学位。主要研究方向为高性能光电探测器（PD），在光互连、消费电子等领域的广泛应用以及基于铌酸锂的集成光子学器件。

（五）

报告时间：16:10-16:30

报告人：罗 毅

报告题目：基于LBD-Tikhonov正则化的平移变化系统表面电荷反演算法

主要内容：

The surface charge accumulation at a gas-solid interface distorts local electric field and in turn leads to surface flashover, which restricts the development and application of HVDC GIS/GIL. Surface charge inversion is the premise of studying surface charge accumulation, because it provides a tool to obtain surface charge distribution after measuring surface potential distribution. On the basis of an iterative regularization technique, a surface charge inversion algorithm based on hybrid Lanczos Bi-diagonalization (LBD) and Tikhonov regularization was constructed for a shift-variant system. The least squares problem for charge inversion was projected to a krylov subspace with smaller dimension by LBD method, and Tikhonov regularization was applied to solve the projected problem at each iteration. Moreover, the adaptive weighted generalized cross validation (A-WGCV) method was adopted to choose the regularization parameters automatically for the projected problem. The implementation progress and performance of the hybrid LBD-Tikhonov algorithm were discussed in detail with the simulated examples, and the efficiency and the accuracy were compared with the charge simulation method and Wiener filter method. At last, the effectiveness and reliability of the Algorithm were verified by the experiment and dust figure. This methodology improves the computational efficiency of surface charge inversion and the selection of regularization parameters, which is beneficial to the development of surface charge measurement technology.

报告人简介：罗毅，博士，2016年毕业于重庆大学电气工程及其自动化专业，2016年至今就读于武汉大学电气与自动化学院电气工程专业。主要致力于直流电场下气-固绝缘界面电荷积聚特性及抑制方法研究。针对表面电荷反演计算中不适定问题，提出了基于Lanczos双对角化-Tikhonov正则化的混合表面电荷反演算法，基于自适应的权重GCV函数选择正则化参数，能显著提高反演精度和计算效率；通过实验和仿真模型研究了表面电荷积聚主导机理的转变现象；提出了基于人工蜂群算法以及表面电荷仿真模型的盆式绝缘子优化模型，构建了基于容性电场分布和阻性电场分布的最小化目标函数，抑制了表面电荷积聚，并提高了沿面闪络特性。

（六）

报告时间：16:30-16:50

报告人：郭 喆

报告题目：用于三维磁场探测和神经形态计算的自旋电子器件研究

主要内容：

The era of Internet of Things (IoT) and artificial intelligence (AI) has proposed high requirements on energy-efficient electronic devices in the applications of signal sensing, data storage and computing. Spintronic devices have been regarded as promising candidates, showing great advantages such as nonvolatility, high speed and endurance, and CMOS fabrication technology compatibility. In this report, we focus on the emerging spintronic devices for 3D magnetic field sensing and neuromorphic computing.  Based on the spin-orbit-torque driven domain nucleation and the symmetrical response to the different stimulations, a single device for sensing 3D magnetic fields has been realized, offering high planar integration capability. Meanwhile, compared with the domain wall motion, the sensor based on the domain nucleation mechnism shows better performance of lineartiy which is a key metric for magnetic sensors. On the other hand, the memristive behaviors of spintronic devices have been investigated and exploited to imitate the biological synapse. An artificial neural network based on such spin-synapse has also been constructed for pattern recognition. Finally, motivated by the sensing and in-memory computing functionalities of spintronic devices, we are exploring the integration system of sensing, memory and computing.

报告人简介：郭喆，博士，2012年毕业于东南大学电子科学与技术专业，2018年获得华中科技大学微电子学与固体电子学专业博士学位。主要致力于后摩尔时代低功耗电子器件及其在新兴领域的应用研究，主要包括自旋电子器件和铁电电子器件等。在研究工作中，利用自旋轨道力矩作用翻转磁矩及其对称性分析，实现了单器件测量三维磁场，解决了传统方法无法实现高度平面集成的难题；利用自旋电子器件的忆阻特性和随机特性，分别构建了人工神经网络和物理不可克隆函数，在图形识别和信息安全领域具有重要应用；利用铁电畴驱动纳米裂纹的快速陡峭翻转，制备了存算一体的低功耗铁电器件以及突破传统玻尔兹曼极限的超低亚阈值摆幅晶体管。相关成果发表在Nano Energy，Nano Letters，Advanced Electronic Materials, IEEE EDL, Applied Physics Letters等期刊上。获得3项基金支持，包括青年基金、博士后特别资助和面上资助。

（七）

报告时间：16:50-17:10

报告人：龚若涵

报告题目：深度学习在电磁场计算中的应用

主要内容：

Convolutional neural networks (CNN) have shown great potentials and have been proven to be an effective tool for some image-based deep learning tasks in the field of computational electromagnetism (CEM). FEM computation results are represented in Red-Green-Blue (RGB) images and the DL prediction outputs are evaluated by mean absolute error (MAE). The possibility and feasibility of the proposed approach are investigated by discussing the influence of various network parameters. It is shown that DL based on U-net can be used as an efficiency tool in multi-physics analysis and achieve good performance with only small datasets. The efficiency of DL mainly depends on how many training samples are needed to effectively converge the network. The sample preparation process often involves a lot of numerical computations, which can be very expensive and time-consuming. In this paper, based on the traditional DL network training procedure, two different approaches, namely adding smart training samples and reference samples, are proposed to help the DL network converge. In addition, the methodology of energy minimization is integrated into the CNN by introducing the physics-informed loss function. It is observed the introduction of the physics-informed loss functions improved the accuracy of the network with the same architecture and database.

报告人简介：龚若涵，博士，2012年毕业于武汉大学电子科学与技术专业，2018年获得武汉大学高电压与绝缘技术专业博士学位。针对装备快速优化设计和复杂电磁场计算等电工领域的热点难点问题，主要从事电工装备多物理场分析计算及深度学习应用的前沿学科研究，尤其在基于数据驱动的替代模型、深度学习的物理信息导入等方面取得了一系列研究成果。构建了面向电工装备的小样本学习深度网络，并通过引入先验物理信息和其他额外约束，开发了针对磁场、温度场等物理信息分布场的深度学习模型，取得了良好的应用效果，相关成果近年来发表了多篇相关领域高水平国际期刊论文。