基于深层玻尔兹曼机的风电场异常风速值自适应检测预处理方法

Adaptive Detection and Preprocessing Method for Abnormal Wind Speed of Wind Farm Based on Deep Boltzmann Machine

为提高现有风电场数据采集系统的准确性和可利用性,提出了深层玻尔兹曼机(DBM)、经验模态分解(EMD)和隐马尔科夫(HMM)组合算法处理运行数据含有少量异常风速值的自适应检测方法。针对风速序列的随机多变性,采用DBM预测方法挖掘异常风速值的潜在特征,得到反映风速值异常情况的残差序列;进一步提高检测精度和降低系统误差的干扰,采用EMD方法捕获残差序列中粗大误差的特征;借助HMM算法的双重随机过程自适应地并剔除检测异常风速点,避免了传统阈值检测方法难以准确识别异常值的问题;最后,为了得到完整的风速序列,对检测出的异常点运用加权双向ARMA算法修正数据。RBF预测结果验证表明,经预处理后风速质量得到了提高,所提方法与传统小波异常值检测方法相比具有更精确的辨识能力,进一步提高了短期风速的预测精度。

To improve the availability and accuracy of data acquisition system of existing wind power plant,this study puts forward the adaptive detection pretreatment method of abnormal wind speed value based on the deep Boltzmann machine(DBM),empirical mode decomposition(EMD)and hidden Markov model(HMM)combination algorithm.Due to the random variability of wind speed sequences,the DBM prediction method is adopted to excavate the potential characteristics of abnormal wind speed value,and get the residual sequences reflecting the anomaly wind speed value.In order to further improve the detection accuracy and reduce the system error interference,the EMD method is adopted to capture the characteristics of bulky errors of the residual sequences.With the help of the Dual stochastic process of HMM algorithm,the abnormal wind speed points are adaptively detected and eliminated,thereby avoid difficulty in accurate outlier identification of the traditional threshold detection method.Finally,in order to get a complete sequence of wind speed,weighted bi-directional ARMA algorithm is taken to revise the data of detected abnormal points.RBF prediction results verify that preprocessing can improve the quality of wind speed.The proposed method,compared with traditional wavelet outlier detection method,is more accurate in identification and further improves the prediction accuracy of short-term wind speed.