“A Neural Network Ensemble-Based Appliance Identification Method for Non-Intrusive Load Monitoring,” Zhaoyuan Fang
Abstract: Approximately 40% of U.S. energy consumption and greenhouse gas emission is contributed by buildings in the residential and commercial sectors. Via changes in user habit and upgrades of appliances, 20% of this could be avoided. With the advance in smart meter installation around the globe, non-intrusive load monitoring (NILM), also known as load disaggregation, is one of the research efforts that have drawn the most attention. NILM refers to the computational or analytical technique to derive the energy consumption of individual appliances from an aggregated energy measurement of the whole household. Compared with intrusive load monitoring, NILM techniques are cost-effective and deployable in real-life applications.
In this work, we propose an NILM method to tackle the load identification problem. Individual appliance power signals from sub-meters and aggregate active power signal are used for an unsupervised event detection and sample generation procedure and a supervised neural network ensemble training. Experiments on the reference energy disaggregation dataset are performed to show that our method is capable of identifying load type and status with high overall Fmeasure.
Bio: Zhaoyuan “Andy” Fang is a rising junior majoring in electrical engineering at the University of Notre Dame.
“Impact of Distributed Energy Resources on Bulk Electric System,” Amrithagunaraj Yogarathinam
Abstract: Conventionally, for system planning and operational studies, the bulk electric system (BES) has been represented by a mesh network connecting equivalent generators and static passive load. The equivalent generators are representatives of central power plants and the static passive load are representatives of the distribution system. Future grids with high penetration of distribution energy resources (DERs) are expected to significantly impact the power flows between the transmission network and the distribution system. In these scenarios, it is recommended to model the DERs in both BES steady-state and dynamic studies. Modeling distribution systems with a high penetration of DERs as a single static load is insufficient as this model fails to represent the DER’s dynamic characteristics and the effects of the unbalanced system. Although, it is desirable to model the distribution system and the multiple DERs connected at various nodes in the distribution system as thoroughly as possible; the computational burden and complexity of such a model may limit the type of studies performed. To that end, this talk will discuss the approaches of modeling the distribution system and its integrated DERs to study the impact of DERs connected to distribution system on the BES. The observations made by comparing different approaches using dynamic time-domain simulations in MATLAB/Simulink platform will be presented. Future research directions will also be discussed.
Bio: Amirthagunaraj “Raj” Yogarathinam is a Ph.D. candidate in the School of Electrical Engineering and Computer Science at the Pennsylvania State University. He received his B.Sc. degree in electrical and electronic engineering from the University of Peradeniya, Sri Lanka, in 2013. His research interests include power system dynamics and control, power electronic control applications in renewable integration to modern power grid, HVDC, multi-terminal HVDC, hybrid- multi-terminal dc, wide-area monitoring and control, dynamic security assessment of power system, and power system economics and optimization.
“Spatial-Temporal Wind Power Probabilistic Forecast with Quantile Regression,” Bo Wang
Abstract: Although wind energy could bring a lot of advantages, many challenges have also been introduced due to the uncertain and intermittent nature of wind. To support both long-term planning or short-term operational decisions, we need to construct an accurate and efficient wind power forecast model. Most of the existing literature focuses on localized wind power or wind speed prediction, based on historical data or meteorological information collected onsite. However, when historical data of various locations are available, the spreading nature of wind suggests a spatial-temporal correlation among different sites, which needs to be considered in the forecast model. This is a common case while predicting future power generation for multiple wind farms simultaneously to guide system-level decisions. In addition, incorporating spatial-temporal dependence would allow us to do inference at potential site without sensors installed or historical data collected.
In this work, we propose a spatial-temporal quantileregression model. Specifically, at any time and location, the wind power generation uncertainty is characterized by a sequence of quantiles. We develop a Gaussian process based meta-model to simultaneously model a sequence of quantiles with the spatial and temporal dependence. A Bayesian inference framework with Markov Chain Monte Carlo sampling procedure will also be discussed.
Bio: Bo Wang is a Ph.D. student in the Department of Industrial and Systems Engineering at Rensselaer Polytechnic Institute. He received his master’s in statistics at Columbia University. His research interests including data analytics, input modeling, output analysis of stochastic simulation, and simulation-based risk management for cyber-physical systems