华电技术 ›› 2019, Vol. 41 ›› Issue (8): 27-31.

• 研究与开发 • 上一篇    下一篇

基于多变量选择的深度神经网络功率曲线建模

  

  1. 华北电力大学 控制与计算机工程学院,北京〓102206

  • 出版日期:2019-08-26 发布日期:2019-09-06

Deep neural network modeling on power curve based on multivariable selection

  1. School of Control and Computer EngineeringNorth China Electric Power UniversityBeijing 102206China

  • Online:2019-08-26 Published:2019-09-06

摘要:

风电机组功率曲线能够反映机组的发电能力,对机组历史数据进行功率曲线建模对风电场的运行管理具有重要意义。引入偏最小二乘法(PLS)在数据层面分析了机组多个变量与输出功率的相关程度,通过交叉有效性原则与投影重要性指标(VIP)对多个变量进行降维筛选,之后把最优的变量子集作为深度神经网络(DNN)的输入,最终得到功率曲线的DNN模型。以安徽某风电场风电机组数据为例,通过与其他建模方法做对比,验证了所提方法的有效性。

关键词:

font-size: 10.5pt, mso-spacerun: 'yes', mso-font-kerning: 1.0000pt">风电机组, 功率曲线, 偏最小二乘法, 深度神经网络, 风电场, 降维筛选

Abstract:

A wind turbine power curve can reflect the generating capacity of a unit. Power curve modeling based on historical data is significant to the operation and management of wind farms.The partial least squares (PLS) method is introduced to analyze the correlation between multiple variables and output power of units on data layer. The crossvalidity principle and the variable importance in  projection (VIP) are used in dimension reduction screening for multiple variables.The subset of optimal variables is used as the input to the deep neural network (DNN), from which the DNN model of the power curve is obtained. Taking the data of a wind turbine in an Anhui wind farm as an example, the effectiveness of the proposed method is verified by comparing it with other modeling methods.

Key words:

font-size: 10.5pt, mso-spacerun: 'yes', mso-font-kerning: 1.0000pt">wind turbine, power curve, partial least squares, DNN, wind farm, dimension reduction screeningfont-size: 10.5pt, mso-spacerun: 'yes', mso-font-kerning: 1.0000pt">