Real-Time Electric Power Telecontrol Transmission Anomaly Detection Based On Data Mining

Power telecontrol transmission anomaly detection can ensure the stable operation of the power grid, reduce or even avoid the loss caused by the anomaly of the power grid. The main challenge of the power telecontrol transmission anomaly detection is how to extract useful information from large amounts of telecontrol transmission data and to design effective data mining model so as to achieve the accurate real-time anomaly detection analysis. This paper focuses on the systematic research of the power telecontrol transmission anomaly detection based on data mining and provides viable ideas for the implementation of the real-time anomaly detection system.This paper firstly introduces the objective and significance of the power telecontrol transmission anomaly detection research and reviews the present research status of the data mining and the anomaly detection. Meanwhile, the features of the power telecontrol transmission system are also introduced and analyzed. Based on these points, extensive researches are conducted.In the step of the data preprocessing for the communication data of the power telecontrol transmission system, we use wavelet transform to analyze the frequency domain of the communication signals and extract the features of the signals. With the features, we realize the initial anomaly detection. At the same time, these features are also used as characteristic properties of the signals for the further data mining analysis.Taking the characteristics of the power telecontrol transmission into consideration, this paper proposes two double-level data stream clustering algorithms for the anomaly detection. The first algorithm, MClu Stream algorithm, a double-level clustering algorithm based on Mahalanobis distance, integrates K-means clustering algorithm and CURE clustering algorithm and uses Mahalanobis distance instead of Euclidean distance to calculate the similarity of the clusters. The second algorithm, DBClu Stream algorithm, is a double-level clustering algorithm based on density grid. It improves the stability of K-means algorithm and reduces iteration time by using the density grid to initialize the cluster centers. Meanwhile, the improved DENCLUE algorithm can achieve a good analysis of the aspheric-distributed data sets with smaller computational complexity. To verify the effectiveness and feasibility of the algorithms, this paper makes the simulation and the analysis for both the algorithms. Compared with classical algorithms, the proposed algorithms get higher detection rate, lower false positive rate and greater stability.After getting the results of wavelet transform analysis and double-level clustering algorithm, this paper adapts Bayesian algorithm to analyze the results and improve the accuracy of the anomaly detection system. What’s more, this paper verifies the effectiveness of the algorithm with the actual data of power telecontrol transmission system. In the end, this paper use the real-time computing framework, Storm, to build the real-time anomaly detection system based on Bayesian algorithm and optimize the system with the distributed computing framework, Hadoop, by improving system concurrency and reducing the computational delay.