Generative Wind Power Curve Modeling Via Machine Vision : A Deep Convolutional Network Method with Data-Synthesis-Informed-Training
Research output: Journal Publications and Reviews (RGC: 21, 22, 62) › 21_Publication in refereed journal › peer-review
Author(s)
Related Research Unit(s)
Detail(s)
| Original language | English |
|---|---|
| Journal / Publication | IEEE Transactions on Power Systems |
| Publication status | Online published - 5 May 2022 |
Link(s)
Abstract
This study develops a novel data-synthesis-informedtraining U-net (DITU-net) based method to automate the wind
power curve (WPC) modeling without data pre-processing. The
proposed data-synthesis-informed-training (DIT) process has
following steps. First, different from traditional studies regarding
the WPC modeling as a curve fitting problem, we renovate the
WPC modeling formulation from a machine vision aspect. To
develop sufficiently diversified training samples, we synthesize
supervisory control and data acquisition (SCADA) WPC data
based on a set of S-shape functions representing WPCs. These
synthesized SCADA data and WPC functions are visualized as
images, named the synthesized SCADA WPC and synthesized neat
WPC, and paired as training samples. A deep generative model
based on U-net is developed to approximate the projection
recovering the synthesized neat WPC from the synthesized
SCADA WPC. The developed U-net based model is applied into
observed SCADA data and can successfully generate the neat
WPC. Moreover, a pixel mapping and correction process is
developed to derive a mathematical form depicting the neat WPC
generated previously. The proposed DITU-net only needs to train
once and does not require any data preprocessing in applications.
Numerical experiments based on 76 WTs are conducted to validate
the superiority of the proposed method via benchmarking against
classical WPC modeling methods. To enable repeating presented
research, we release our code and high resolution figures at
https://github.com/IkeYang/DITU-net.
power curve (WPC) modeling without data pre-processing. The
proposed data-synthesis-informed-training (DIT) process has
following steps. First, different from traditional studies regarding
the WPC modeling as a curve fitting problem, we renovate the
WPC modeling formulation from a machine vision aspect. To
develop sufficiently diversified training samples, we synthesize
supervisory control and data acquisition (SCADA) WPC data
based on a set of S-shape functions representing WPCs. These
synthesized SCADA data and WPC functions are visualized as
images, named the synthesized SCADA WPC and synthesized neat
WPC, and paired as training samples. A deep generative model
based on U-net is developed to approximate the projection
recovering the synthesized neat WPC from the synthesized
SCADA WPC. The developed U-net based model is applied into
observed SCADA data and can successfully generate the neat
WPC. Moreover, a pixel mapping and correction process is
developed to derive a mathematical form depicting the neat WPC
generated previously. The proposed DITU-net only needs to train
once and does not require any data preprocessing in applications.
Numerical experiments based on 76 WTs are conducted to validate
the superiority of the proposed method via benchmarking against
classical WPC modeling methods. To enable repeating presented
research, we release our code and high resolution figures at
https://github.com/IkeYang/DITU-net.
Research Area(s)
- Adaptation models, Data models, data-driven model ACRONYMS 4PLF Four-parameter logistic function 5PLF Five-parameter logistic function ADE Adjusted double exponential BSA Backtracking search algorithm CDF Cumulative distribution function CWS Cut-in wind speed DCAE Deep convolutional autoencoder DE Double exponential DIT Data-synthesis-informed-training DITDCAE Data-synthesis-informed-training deep convolutional autoencoder DITU-net Data-synthesis-informed-training U-net DKC Domain knowledge correction IDP Insufficient data pattern KNN K nearest neighbor MAE Mean absolute error MAPE Mean absolute percentage error ML Machine learning MS Marker size, Machine vision, Mathematical models, neural networks, Training, wind energy, wind power curve, Wind power generation, Wind speed, wind turbine
Citation Format(s)
Generative Wind Power Curve Modeling Via Machine Vision : A Deep Convolutional Network Method with Data-Synthesis-Informed-Training. / Yang, Luoxiao; Wang, Long; Zhang, Zijun.
In: IEEE Transactions on Power Systems, 05.05.2022.Research output: Journal Publications and Reviews (RGC: 21, 22, 62) › 21_Publication in refereed journal › peer-review
