• WeChat official account for the department
    WeChat official account for undergraduate students of the department
    WeChat official account for graduate students of the department
    WeChat official account for alumni of the department
    WeChat official account for the Energy Internet Research Institute, Tsinghua University
    WeChat official account for the Sichuan Energy Internet Research Institute, Tsinghua University







Current Location: Home > News > Notice > Content

Topic:Electricity Market Forecasting via Generative AI

 Time:May 10, 10-12am

Add:West Main Building, 3-102

 Reporter:Prof. Lang Tong, Cornell University

Bio: Lang Tong is the Irwin and Joan Jacob Professor of Engineering at Cornell University and the Site Director of the Power System Engineering Research Center. His current research focuses on power system optimizations, power economics and electricity markets, and data analytics, machine learning, and AI technologies for power system operations. He received a B.E. degree from Tsinghua University and a Ph.D. from the University of Notre Dame. A Fellow of IEEE, he was the 2018 Fulbright Distinguished Chair in Alternative Energy.


Abstract:A defining feature of generative artificial intelligence (AI) is its ability to produce artificial samples that resemble reality. This talk presents a novel generative probabilistic forecasting approach derived from the Wiener-Kallianpur innovation representation of nonparametric time series. In particular, we propose a weak innovation autoencoder architecture that transforms nonparametric multivariate random processes into canonical innovation sequences, from which future time series samples are generated according to conditional probability distribution on past samples. A novel deep-learning algorithm is proposed that constrains the latent process to be an independent and identically distributed sequence with matching input-output probability distributions of the autoencoder. Three applications involving highly dynamic and volatile time series in electricity markets are considered: (i) locational marginal price forecasting for merchant storage participants, (ii) price spread forecasting for virtual bidding in interchange markets, and (iii) area control error forecasting for frequency regulations. We compare the proposed innovation-based forecasting with classic and leading machine-learning techniques, including some of the large language model-based forecasting techniques.

—— Share ——

Previous:2024 Summer School for Future Power and Energy System (FPES)

Next:Jingde Podium· Preview | Innovative Industrial Cooperation in Electrical Engineering – Opportunities and Challenges