Presenter Profile
REMOTE PRESENTER
Yinghui Wu
Associate Professor
Case Western Reserve University, Department of Computer and Data Sciences, School of Engineering
Yinghui Wu is an Associate Professor in the Department of Computer and Data Science in the Case Western Reserve University. He holds a joint position as a staff scientist at the Pacific Northwest National Laboratory. His area is in Data and Knowledge management, Databases, Data Mining and Machine Learning. He received his Ph.D. in Computer Science from the University of Edinburgh, UK, and B.S. in Computer Science from Peking University, China. His area is in data management and analytics, including data quality, graph query processing, and knowledge base systems. His current research develops scalable graph analytical systems for multidisciplinary database applications. He serves as an associate editor for ACM Journal of Data and Information Quality. He also currently co-directs the Materials Data Science Center of Excellence (MDS3) with Dr. Roger French.
TALK TITLE
Graph-based Deep Learning for PV Performance Degradation Analysis
KEYWORDS
Photovoltaic (PV) power stations, PV Performance Loss Rate (PLR), Graph deep learning techniques (Graph Neural Networks - GNNs), Spatio-Temporal Denoising Graph Autoencoder (STD-GAE), Spatio-temporal Dynamic Graph Neural Network (ST-DynGNN)
TALK ABSTRACT
Photovoltaic (PV) power stations have become an integral component to a sustainable energy production landscape. Accurately estimating and predicting performance of PV systems is critical to their feasibility as a power generation technology and as a financial asset. Among the major domain problems is to understand the PV Performance Loss Rate (PLR) for large fleets of PV systems. The integration of the global PV market with real time data-loggers has enabled large scale PV data analytical pipelines for power forecasting and long-term reliability assessment of PV fleets. Nevertheless, such analysis heavily depends on the quality of PV data. We present our recent work on exploiting and innovating graph deep learning techniques (Graph Neural Networks - GNNs), that address two major challenges: (1) missing PV data imputation, and (2) diversity of degradation patterns from PV data to support “end-to-end” long-term PV performance degradation analysis. Specifically, for (1) we introduce a Spatio-Temporal Denoising Graph Autoencoder (STD-GAE) framework to impute missing PV Power Data. STD-GAE exploits temporal correlation, spatial coherence, and value dependencies from domain knowledge to recover missing data. For (2), we outline a novel Spatio-temporal Dynamic Graph Neural Network (ST-DynGNN) that adopts a paralleled graph autoencoder architecture to extract different aging and fluctuation terms simultaneously. Experimental results show that, compared with state-of-the-art counterparts, STD-GAE can achieve a gain of 43.14% in imputation accuracy and remains less sensitive to missing rate, different seasons, and missing scenarios; and our preliminary results on ST-DynGNN obtains a reduction of 34% on average in MAPE in PLR degradation analysis.