Automatic Analysis Substation Monitoring Background Alarm Signals Based On Data Mining

This study applies data mining to the alarm signal analysis in substation monitoring systems. We add hyperlinks to the alarm light indicators and present all possible failure nodes which maytrigger the alarm signal to the operators on duty. When the monitoring system sends an alarm signal, the corresponding alarm light indicator is turned on, the operators can get prompt messages information of all possible failure nodes with respect to that alarm signal by just clicking the light indicator. This technique can help the operators save the time to look through the relevant technical documents and equipment drawings. It changes the current situation that the analysis of monitoring alarm signal is very dependent on the personal experience of the operators, and avoid the failure expansion caused by failure misjudgment or decision delay due to unusual alarm signals, missing or damaged equipment drawings, etc.So this technique can help the operatorshandle the failure of electrical equipment correctly and timely.This study selects several typical alarm signals to test. We verify the feasibility to classify and merge the alarm signals. For the alarm signals generated by the power equipment from the same manufacturer, the alarm signals of the same type are merged together, which alleviates the work a lot to add prompt message hyperlinks for each alarm signal. More importantly, the classification and merging of the alarm signals make the loop of each alarm signal complete, which ensures this technique adds the prompt message for the alarm signal without missing any possible failure node.When performing data mining on the maintenance records in the current power grid production management system, we find that the query display of currentsystem has some serious defects which affect the efficiency heavily. For example, the system can not subdivide the equipment manufacturers of interested and can not display the conclusion of maintenance records directly. We repair these defectsto get more useful failure node information while at the same time improve the efficiency of data mining significantly, which lays a solid foundation for the failure node sorting in the next step.In this study, when the substation monitoring system gives prompted messages for each alarm signal, it requires not only the messages show comprehensive and complete information of node failure, but also sort these nodes according to their possibility of failure. This study draws on the idea of "shopping cart"of supermarket to analyze which productsare often also appear in the customer’s shopping cart simultaneously, in order to place the products which will most likely to be purchased together on the adjacent shelves. The study applies association FP-growthmining algorithm to dig out the alarm signals and the failure nodes which appear in the maintenance records simultaneously, and sorts them according to the probability of occurrence of failure nodes and certain alarm signal, i.e., support filtering and confidence sorting, to achieve the ranking of failure nodes appeared in the alarm prompt message.Taking into account that new records are written into the database of the grid production management system almost every day, this studyapplies the incremental FSPM-FP mining algorithm on the new maintenancerecords to dig out the frequent nodes first, and then divides the sorting data mining of overall database into several situations. By making full use of the FP-tree constructed in the initial sorting, we reduce the total number of database traversal, and get an efficient algorithm and the sorting result which is closest to the actual situation. This study can provide real guidance to the operators when analyzing the alarm signals, reduce the loss caused by equipment failure to a minimum, and guarantees the stable power grid and the electricity safety.