A Case-Study of Wind Turbine Power Forecasting Using Machine Learning Techniques
(Room Silk 3)
20 Sep 17
2:00 PM
-
3:30 PM
Tracks:
Track G - Renewable Energy Technology
Speaker(s):
Chandrasekar Venkatraman, Principal Research Scientist, Hitachi America, Research and Development, Big Data Laboratory, Santa Clara, CA.; Pierre Huyn, Principal Research Architect, Hitachi America, Research and Development, Big Data Laboratory, Santa Clara, CA.; Panitarn Chongfuangprinya, Principal Research Scientist, Hitachi America Ltd. ; Sunder Raju, Managing Director, Atria Power Corporation
Renewable energy resources such as wind generation are enjoying a growing proliferation driven by a mixture of favorable legislation, incentives, technology developments and cost reduction. These resources enable utilities companies to provide cheaper and cleaner services, and new levels of customer satisfaction. However, due to its intermittent nature and higher variability, increasing penetration of wind and solar power generation leads to potential impacts on planning and operations of power systems. Power generation forecasting for renewable energy sources has become an important requirement in the energy industry. This paper presents a case study of power generation forecasting for a wind turbine farm operated by a major Independent Power Producer (IPP) based in India. Like many others, this IPP is mandated to provide accurate day-ahead power generation forecasts in 15-minute intervals. In this case study, we propose an Internet of Things (IoT) platform for the energy domain, through which SCADA data is acquired from wind turbines in real time, cleansed, aggregated, compressed, and securely transmitted to the cloud for analysis. To produce forecasts, we developed forecast models using Machine Learning techniques that can learn and execute large numbers of models efficiently and provide accurate predictions. Efficient use of computing resources in forecasting proves to be important as increasing numbers of forecast analytics will be deployed to edge computing devices. We also address other challenges such as when to refresh the forecast models, how to integrate external meteorological data, tradeoffs between data acquisition sampling rates and forecast accuracy, and how to improve reliability of our IoT pipeline. The proposed solution can be used for any renewable energy system such as wind generation, Photovoltaic Distributed Generation (PV-DG) system, and hybrid systems that combine wind generation and PV-DG.