PCIM研讨会 | 特邀合工大、华科大、电子科大、卓迈共探 AI 赋能电力电子自动化设计

"AI 不再是行业噱头，而是颠覆电力电子传统试错研发模式、实现降本增效的核心生产力。"

从局部优化到通用设计平台——电力电子变换器及部件的智能化实践与展示

演讲简介：

Artificial intelligence is not a fleeting hype, but a tsunami that is fundamentally reshaping every industry — and power electronics is no exception. The key pain points in today’s industry are clear: converter and core component design still heavily relies on expert heuristics and costly trial and error cycles, leading to long development times and high costs. Meanwhile, generic large language models (LLMs) lack access to high quality design data and thus cannot deliver truly optimal solutions that meet specific customer requirements. Our proposal is an agent based system — a feasible path toward fully automated power electronics design.This lecture will begin with a concise review of recent AI advances in power electronic converter and component design. It will then focus on the author’s nearly ten years of hands on “AI + power electronics” practice, using electric vehicle on board chargers (OBCs) and power module packaging as concrete examples. You will see how AI can be applied to open loop, modulation, control, packaging, multi-physics design etc — delivering measurable improvements in design efficiency.Worth mentioning that: in the final part of the lecture, we will move beyond isolated AI applications and present — for the first time at PCIM Asia — a live demonstration of a general AI design platform for power electronic converters and components. The demo will show an automated design flow from user requirements to a generated design solution.

演讲人简介：

Professor, Doctoral Supervisor, Delta Young Scholar, recipient of the Anhui Provincial Outstanding Youth Fund Talent Program, and Hefei Most Beautiful Science and Technology Worker. He received his Ph.D. from Delft University of Technology (the Netherlands), his Master's degree from Xi'an Jiaotong University, and his Bachelor's degree from Southeast University. He has led research projects with a total funding of nearly 60 million RMB, published over 100 papers, and holds more than 50 patents. He is the lead author of two monographs and a co-author of one. He has established the first power semiconductor packaging and reliability testing platform in Anhui Province, reaching advanced domestic and international standards. His research achievements have been implemented in several companies, including Sungrow. He has received the First Prize of Anhui Provincial Science and Technology Progress Award, the Grand Prize of Science and Technology Progress from the China Power Supply Society, among other honors. His main research interests include AI-driven power electronics design, power semiconductor packaging and integration, and interdisciplinary studies in power electronics.

电力电子设计智能化：从智能算法到智能系统

演讲简介：

当前，人工智能技术正经历着从"简单AI算法"、"对话式AI"和"推理式AI"向"智能体"的深刻转变。深入理解AI的内在原理与发展轨迹，并将其融入电力电子设计的各个环节，是一个值得深入探讨的课题。本报告将结合讲者在电力电子设计领域的实践经验，分享相关思考与探索，以期为同行提供参考。    

演讲人简介：

陈宇，教授，博导，国家级青年人才项目入选者，湖北省首届青年拔尖人才项目入选者。研究领域为电力电子标准化和智能化设计。获国家科技进步二等奖，中国电源学会优秀青年奖，全国百篇优秀博士论文提名奖。首批国家级一流线上课程《电力电子学》的主讲教师，《电力电子学——电力电子变换和控制技术（第四版）》主编，《电力电子技术（微课版）》主编。    

AI-Enabled Optimization of Modulation Techniques in Power Electronics

演讲简介：

Propelled by the rapid evolution of next-generation artificial intelligence (AI), deep reinforcement learning (DRL) has proven highly effective in tackling complex, high-dimensional optimization problems under uncertainty. This presentation will share practical insights and deployment strategies for applying these advanced techniques across three key domains of power electronics.We will explore the shift beyond traditional methods to truly intelligent design frameworks, outlining a clear roadmap for harnessing AI to revolutionize industrial power electronics.

演讲人简介：

胡维昊，电子科技大学教授、博导、IET Fellow，曾入选国家青年特聘专家，四川省学术技术带头人，四川省杰青等人才计划。主要从事新型电力系统与人工智能技术交叉领域研究。主持了由国家重点研发计划、国家自然科学基金、四川省科技厅等资助的科研项目10余项。以第一/通讯作者身份发表SCI论文150余篇，论文被引用16000余次，H因子为68，获IEEE 国际会议最佳论文奖6次，授权发明专利30余项。撰写国内第一本人工智能与电气应用交叉学科方向的专著《人工智能与电气应用》。获中国仪器仪表学会科学技术一等奖1项和教育部科技进步二等奖1项。担任智能电网国家科技重大专项2025年度项目责任专家、IET Fellow评审委员会成员、中国电工技术学会人工智能与电气应用专业委员会主任委员等学术职务。    

Understanding, Not Just Search — Engineer-Experience-Driven Vertical Agents for Semiconductors

演讲简介：

This talk addresses one core question: **How can conversational AI move from "looks right" to "engineering-verifiably right" in professional domains?**We first review five eras of conversational AI — Rule-based, NLU, RAG, Agent, and Vertical Agent — analyzing trade-offs in the *accuracy—coverage—controllability* triangle, and identify three fundamental gaps in general-purpose LLMs: outdated data freshness, lack of substitution-relation modeling, and absence of online-offline integration.**Central topic: How do we systematically embed 20 years of hardware engineering experience into an agent?** We unpack three layers:- **Knowledge Graph:** Extracting chips, parameters, and topologies from datasheets, together with substitution / compatibility / constraint relations — encoding a four-tier substitution standard and PPN/OPN dual-layer versioning.- **Intent Recognition:** *"Can NCP1342 design a 100W supply?"* and *"What is NCP1342's max duty cycle?"* look similar but execute through completely different toolchains — we design 8 primary intents × multi-dimensional slots for structured extraction.- **Business Orchestration:** A Planner dispatches 15+ tools, each step grounded on the previous tool return, confidence monotonically non-increasing, every parameter traceable — the essence of *Separation of Understanding & Control*.**The highlight is a live end-to-end demo.** A single natural-language request — *"Design a 100W flyback supply, 90–264VAC input"* — drives Lumy through intent recognition → topology recommendation → controller chip selection → peripheral component calculation (turn ratio, sense resistor, compensation network) → substitution analysis → final output: a PCB-ready BOM and reference schematic. **One sentence in, one engineering drawing out.**We close with the broader significance of vertical agents: converting tacit engineering know-how into transferable structured assets in the face of an aging workforce, with a methodology directly transferable to medicine, law, and finance.**Contribution:** We propose and validate a reference architecture for vertical agents with deeply embedded engineering experience, and present — for the first time — a complete end-to-end reasoning chain from a design idea to an engineering drawing.

演讲人简介：

利兹大学AI博士，多篇顶会顶刊作者，IEEE TITS审稿人。在国际上率先提出“自适应不确定性量化”预测框架，突破传统模型可靠性瓶颈；主导研发国内首个面向旅游规划的“时空增强大模型”系统，实现复杂因素与知识图谱的高效交互，显著提升回答一致性。

扫码锁定席位

扫码预览议程

浙江大学、英飞凌科技、

香港微电子研究院

最佳论文奖 | 青年工程师奖

高校科学家奖 | 优秀墙报奖

一位长期活跃在全球电力电子行业

的PCIM研讨会主席

二十年行业洞察：透视功率半导体未来，

为何PCIM Asia是不可或缺的支点？