Research On Data Mining Technology In Intelligent Substation Information Integration Platform

According to the comprehensive development of intelligent substations, the request of information interaction between master substations and substations, and data access from advanced applications grows dramatically. Based on network technologies, information platform of substation provides services for various data access applications. Data mining is one of the most magnificant techniques to analyze and process massive structural and unstructural data in substations, which is applied widly in prediction, association, clustering, classification, time series analysis, etc. In this dissertation, we focus on relative algorithms of data mining embedded on substation information platform in the perspective of intrusion detection for communication isolating device, image segmentation and recognition for intelligent inspection system, and target detection from the monitoring video stream. The primary content includes followings:1. To improve the intrusion detectionpart of the isolating devices, a dimension reduction algorithm is proposed based on invariant moment technique, and an improved ART2-SVM model is applied to achieve classification. For reliable running and defence ability improving purpose, an improved detection model is proposed to fix the problem of low detection rate and poor dimension reduction ratio. In this model, we extract invariant moment features of intrusion records for dimension reduction, and then construct a classifier composed of a transformed ART2 network and a SVM model for machine learning and the ultimate classification. In the experimental part, KDD 99 dataset is used as the simulated hack records, and the result indicates that this proposed model for dimension reduction and classification could work effectively to improve the detection rate of the existing isolating devices.2. In the intelligent inspection system part, an improved combined image feature extraction algorithm is proposed based on marked watershed, Krawthouk invariant moment, and GLCM. Futher, an improved DENCLUE algorithm is proposed to perform the clustering and pattern recognition based on revised HSIM fuction. While performing inspection tasks in different locations or perspectives, inspection robots often photograph equipment pictures within the case ofoccurrence of obscured objects and affine transformation. Since regular image processing methods are not sufficient to cope with this situation, an improved combined imagefeature extraction model is proposed in this dissertation. First, marked watershed method is utilized to segment equipment objects in the original picture, and then Krawthouk invariant moment and GLCM statistics is calculated to synthetize feature vectors for each equipment object. Then, an improved DENCLUE method based on revised HSIM function is applied to realize the clustering, recognition, and outlier detection. The experiment result verifies this algorithm could reform the regular image processing procedure and work properly under the situation of occurrence of obscured objects and affine transformation, and could recognize the abnormal object in the picture by outlier detection.3. In the intelligent surveillance system, an improved Guassian mixture model is introduced for running status detection of substation equipments. According to the current situation, the surveillance system can not report the abnormal running status of the equipments from the video. Though target detection is widly used in intelligent surveillance system, regular detection method is failed to depart background pixels from foreground objects when abnormal vibration occurs with a small amplitude. An improved GMM model is proposed to report the abnormal vibration movement by time series method, which records the movement trail for background pixels from the original GMM model, and decides the attribution of each pixel. This algorithm could provide an reasonal basis for equipment failurediagnosis, intelligent maintenance, and running management in severse weather.4. Auserinterface of data mining system for substation is designed and realized, including the classification, clustering, and time series analysis methods in IDS, intelligent inspection system, and intelligent surveillance system respectively. This userinterface could easily extended, and provide a proper platform for succeeding works.