Research Of WAMS Real-time Data Processing And Anomaly Detection Method For Smart Grid

With the construction of strong smart grid plan proposed,the network structure and the operation model of power grid will become more complex,and the wide area measurement system(WAMS)that based on synchronized phasor measurement unit(PMU)can collect the real-time data from the whole network and the network synchronous angle of each site,and monitor the power grid operation in real-time.Through the analysis of real-time acquisition data,and the found of the anomaly data in time,the effect of the fault can be reduced to a minimum.In this thesis,data acquisition based on the acquisition unit PMU of the wide area measurement system is taken as the research object.Based on the streaming data processing platform Storm and combined with the sliding window technology to finish the anomaly detection of WAMS real-time data.The algorithm based on machine learning has been widely applied in data processing and anomaly detection.And in order to ensure the real-time performance of anomaly detection in WAMS high-dimensional data stream,data dimensionality reduction has become a necessary prerequisite for anomaly detection.First of all,combined with the characteristics of the data from WAMS,this thesis uses dimension function and sliding window reduction technology of auto encoder,can reduce the dimension of the WAMS real-time data flow,designed automatic encoder data stream reduction model based on DSDAE.Then,according to the local outlier factor algorithm has higher time complexity,combined with the K-means clustering algorithm,proposed an improved algorithm named K-LOF local outlier factor density detection algorithm based on LOF,and the optimal neighborhood query process,the experiment results show that K-LOF algorithm can effectively reduce the abnormal to detect the time complexity of the algorithm and improve the detection accuracy.Finally,based on DSDAE data stream reduction model and K-LOF anomaly detection algorithm,the design of WAMS data stream processing and anomaly detection model DSDAEKLOF,and set up the WAMS real-time data stream processing framework in the Storm platform,compared with DSDAE model and single drop dimensional anomaly detection before and after measuring the average delay,verify the validity of the Storm platform based on the data stream processing framework can effectively improve the real-time data processing and auto-encoder data streams based on dimensionality reduction model.