基于滑动窗双边CUSUM算法的风电爬坡事件检测方法  

Wind Ramp Event Detection Method Based on Sliding Window with Bilateral CUSUM

作　　者：冯萧飞 刘韬文 李彬[1] 苏盛[1] FENG Xiao-fei;LIU Tao-wen;LI Bin;SU Sheng(College of Electrical and Information Engineering,Changsha University of Science and Technology,Changsha 410114,China;Hunan Electric Power Corporation,Changsha 410114,China)

机构地区：[1]长沙理工大学电气与信息工程学院,长沙410114 [2]国网湖南省电力有限公司,长沙410114

出　　处：《科学技术与工程》2024年第2期595-603,共9页Science Technology and Engineering

基　　金：国家自然科学基金(51777015)。

摘　　要：随着新能源并网进程的推进,风电装机规模逐年扩大。受区域内天气变化影响,风机出力的间歇性和波动性特征对电网的威胁亦越发显著。极端天气所引发的风电出力异常爬坡事件,易导致电网功率失衡,对电力系统机组调度、源荷平衡造成了极大压力。合理的风电爬坡事件检测以及精准的风电功率预测能为风电场运维及电力系统调度提供先验指导,有力缓解风电不确定性带来的危害。首先讨论了目前主流风电爬坡事件定义的盲点,分类并分析了3种风电爬坡场景的功率变化特性,据此提出基于滑动窗双边累计和(cumulative sum, CUSUM)算法的风电爬坡事件检测方法,提取时序耦合信息,捕捉短时间窗口内风电功率数据的异常波动,提高风电爬坡事件检测精度。其次,采用贝叶斯优化的长短期记忆(long short term memory, LSTM)神经网络,最优化模型超参数,提高模型对于爬坡事件发生时风机出力的预测性能。进一步应用所提风电爬坡事件检测方法,对模型预测区间内的风电爬坡事件进行检测实验,验证了所提方法的有效性。With the advancement of new energy grid integration,the scale of wind power installations has been expanding annually.Affected by regional weather changes,the intermittency and fluctuation characteristics of wind turbine output pose an increasingly significant threat to the power grid.Extreme weather-induced wind power output anomalies,known as ramp-up events,can lead to power grid imbalances and place tremendous pressure on power system unit dispatch and load balancing.Reasonable detection of wind power ramp-up events and accurate wind power forecasting can provide prior guidance for wind farm operation and maintenance and power system dispatch,effectively alleviating the harm caused by wind power uncertainty.Firstly,the blind spots in the current mainstream definitions of wind power ramp-up events are discussed.Following this,the power change characteristics of three wind power ramp-up scenarios were classified and analyzed.Then,a method for detecting wind power ramp-up events based on sliding window bilateral Cumulative Sum(CUSUM)was proposed.This method extracts time series coupling information,captures abnormal fluctuations in wind power data within a short time window,and improves the accuracy of wind power ramp-up event detection.Furthermore,a long short-term memory(LSTM)neural network optimized by Bayesian optimization was employed to optimize model hyperparameters and improve the model s predictive performance for wind turbine output during ramp-up events.The proposed wind power ramp-up event detection method was further applied to detect wind power ramp-up events within the model prediction interval,verifying the effectiveness of the proposed method.

关 键 词：风电爬坡 滑动窗 CUSUM算法 贝叶斯优化 LSTM神经网络 

分 类 号：TM614[电气工程—电力系统及自动化]