BEGIN:VCALENDAR VERSION:2.0 PRODID:IEEE vTools.Events//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:America/New_York BEGIN:DAYLIGHT DTSTART:20260308T030000 TZOFFSETFROM:-0500 TZOFFSETTO:-0400 RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3 TZNAME:EDT END:DAYLIGHT BEGIN:STANDARD DTSTART:20251102T010000 TZOFFSETFROM:-0400 TZOFFSETTO:-0500 RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11 TZNAME:EST END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20251110T053028Z UID:72F60A62-0889-43C4-812D-E6C08E66340A DTSTART;TZID=America/New_York:20251106T120000 DTEND;TZID=America/New_York:20251106T130000 DESCRIPTION:This talk introduces how to address a fundamental challenge in applying\ndeep learning to power systems: developing neural network models that transfer\nacross significant system changes\, including networks wit h entirely different\ntopologies and dimensionalities\, without requiring training data from unseen\nreconfigurations. Despite extensive research\, most ML-based approaches remain\nsystem-specific\, limiting real-world dep loyment. This limitation stems from a dual\nbarrier. First\, topology chan ges shift feature distributions and alter input dimensions\ndue to power f low physics. Second\, reconfigurations redefine output semantics and\ndime nsionality\, requiring models to handle configuration-specific outputs whi le\nmaintaining transferable feature extraction. To overcome this challeng e\, we introduce\na Universal Graph Convolutional Network (UGCN) that achi eves transferability to\nany reconfiguration or variation of existing powe r systems without any prior\nknowledge of new grid topologies or retrainin g during implementation. Our approach\napplies to both transmission and di stribution networks and demonstrates\ngeneralization capability to complet ely unseen system reconfigurations\, such as\nnetwork restructuring and ma jor grid expansions. Experimental results across\ndifferent power system a pplications\, including false data injection detection and state\nforecast ing\, show that UGCN significantly outperforms state-of-the-art methods in \ncross-system zero-shot transferability of new reconfigurations.\n\nCo-sp onsored by: Power Systems Engineering Center (PSEC) at New Jersey Institut e of Technology\n\nSpeaker(s): Dr. Tong Wu\n\n323 Dr Martin Luther King Jr Blvd\, Newark\, New Jersey\, United States\, 07102\, Virtual: https://eve nts.vtools.ieee.org/m/509297 LOCATION:323 Dr Martin Luther King Jr Blvd\, Newark\, New Jersey\, United S tates\, 07102\, Virtual: https://events.vtools.ieee.org/m/509297 ORGANIZER:sangwoo.park@njit.edu SEQUENCE:20 SUMMARY:Universal Grid-Graph Learning: Zero-Shot Transfer without Retrainin g or Fine-tuning URL;VALUE=URI:https://events.vtools.ieee.org/m/509297 X-ALT-DESC:Description: <br /><p>This talk introduces how to address a fund amental challenge in applying<br>deep learning to power systems: developin g neural network models that transfer<br>across significant system changes \, including networks with entirely different<br>topologies and dimensiona lities\, without requiring training data from unseen<br>reconfigurations. Despite extensive research\, most ML-based approaches remain<br>system-spe cific\, limiting real-world deployment. This limitation stems from a dual< br>barrier. First\, topology changes shift feature distributions and alter input dimensions<br>due to power flow physics. Second\, reconfigurations redefine output semantics and<br>dimensionality\, requiring models to hand le configuration-specific outputs while<br>maintaining transferable featur e extraction. To overcome this challenge\, we introduce<br>a Universal Gra ph Convolutional Network (UGCN) that achieves transferability to<br>any re configuration or variation of existing power systems without any prior<br> knowledge of new grid topologies or retraining during implementation. Our approach<br>applies to both transmission and distribution networks and dem onstrates<br>generalization capability to completely unseen system reconfi gurations\, such as<br>network restructuring and major grid expansions. &n bsp\;Experimental results across<br>different power system applications\, including false data injection detection and state<br>forecasting\, show t hat UGCN significantly outperforms state-of-the-art methods in<br>cross-sy stem zero-shot transferability of new reconfigurations.</p> END:VEVENT END:VCALENDAR