一种风电场风速异常数据预处理的新方法  

A new method for data preprocessing of wind speed anomalies at wind farm

作　　者：陈伟[1] 王敏 裴喜平[1] CHEN Wei;WANG Min;PEI Xi-ping(College of Electrical Engineering and Information Engineering, Lanzhou Univ. of Tech., Lanzhou 730050, China)

机构地区：[1]兰州理工大学电气工程与信息工程学院

出　　处：《兰州理工大学学报》2019年第5期91-96,共6页Journal of Lanzhou University of Technology

基　　金：国家自然科学基金(51767017,51867015);甘肃省基础研究创新集体项目(18JR3RA133)

摘　　要：针对风电场采集到的历史风速数据中存在异常值的问题,为保证风速数据的准确性和有效性,提出了一种运用差分自回归滑动平均(auto regressive integrated moving average, ARIMA)、小波分解(wavelet decomposition,WD)和隐马尔科夫(hidden Markov model,HMM)组合算法对异常风速数据进行挖掘的方法.采用ARIMA模型挖掘异常风速数据的潜在特征,得到反映风速值异常情况的残差序列;为进一步提高检测精度和降低系统误差的干扰,采用小波分解方法捕获残差序列中的粗大误差特征;借助HMM算法的双重随机过程检测异常风速值并剔除,将剔除异常值后的数据运用粒子群优化最小二乘支持向量机方法进行重构,保证风速序列的完整性.实际算例结果表明了所提方法的有效性和可行性.Aimed at the existence of abnormal values in historical wind speed data acquired at wind farm and ensuring the data to be accurate and effective, an integral algorithm was proposed to mine abnormal wind speed data by using auto-regressive integrated moving average(ARIMA) of difference, wavelet de composition (WD),and hidden Markov model (HMM). The ARIMA model was used to mine potential characteristics of anomalous wind speed data and a residual error sequence reflecting the anomaly of the wind speed value was obtained. In order to further improve detection accuracy and reduce interference of system error, wavelet decomposition method was used to capture coarse error characteristics in the residu al error sequence. The double random process of the HMM algorithm was used to detect the abnormal wind speed value and it was removed. The data after anomaly rej ecting was reconstructed by using a meth od of particle swarm optimization least squares support vector machine to ensure the integrity of the wind speed sequence. The result of practical example indicated the effectiveness and feasibility of the method proposed in this article.

关 键 词：异常数据识别 时间序列 小波分解 隐马尔科夫模型 

分 类 号：TM614[电气工程—电力系统及自动化]