As a part of a substation-level decision support system, a new data-driven intelligent Hierarchical Fault Diagnosis System (HFDS) for on-line fault diagnosis is presented in this paper. To make the large-scale software system easy to set up and flexible to maintain, the object-oriented programming (OOP) technique is adopted. The physical objects in the T&D system, such as a transmission or a distribution line, a bus bar, or a transformer, are regarded as the objects that belong to diverse classes in the software. The structure of the proposed object-oriented Hierarchical Fault Diagnosis System is shown in Figure 1, which consists of Man-Machine Interface (MMI), Alarm Processing Module (APM), Phase I and Phase II Diagnosis Modules, and Graphical User Interface (GUI). The MMI works as a vital communication medium between the experienced operators and the proposed HFDS. The MMI requires the experienced operators to input the data for the instance variables of each object class by interacting with the experienced operators. These training data are used to set up the APM and the phase I & II diagnosis modules. Based on the raw database, the MMI automatically creates two data-bases, Data Base I and Data Base II, for construction of the APM and the two-phase diagnosis modules. Receiving the time-stamped alarm signals from SCADA interface in on-line environment, the APM is intended to initiate the diagnosis process and compile the alarm data into the format required by the two-phase diagnosis modules. Initiated by the APM, the phase I diagnosis module begins with estimating the possible fault sections according to the alarm data from the SCADA interface, the data which has been processed by the APM. To circumvent the problems of the conventional multilayered artificial neural network in excessive training efforts required, the Group Method of Data Handling (GMDH) network is employed to achieve the same purpose on fault diagnosis, but with much less computational efforts. On the basis of the possible fault sections provided by the phase I diagnosis module and the alarm signals with operating sequence provided by the APM, the phase II diagnosis module accomplishes the fault diagnosis through further identification of the fault type and detailed explanation of the fault situation. To achieve this purpose on-line and real-time, the proposed fast Bit-Operation Logical Inference Mechanism (BOLIM) is employed in the phase II diagnosis (Figure Presented) Figure 1. Structure of the hierarchical fault diagnosis system Table 1. Composition of test cases single fault cases double fault cases triple fault case Category A B A B A B single phase to ground 80 95.0% 80 93-8% 40 92.5% phase to phase fault 45 91.1% 40 92.5% 25 92.0% double phase to ground 45 93.3% 40 95.0% 25 96.0% three-phase short circuit 25 92.0% 20 90.0% 15 93.3% three-phase to ground 25 96.0% 20 95.0% 15 93.3% A is the number of testing cases B is the correct rate of fault type identification. The correct rate of fault section identification for the testing cases reaches as high as 100 percent. module. The BOLIM has the advantage of faster inference than the conventional symbolic processing through the rule base of the expert system, partly because the BOLIM only deals with the judgement of a few alarm signals which are related to the fault sections previously identified in the phase I diagnosis module. Consequently, the proposed fast BOLIM not only improves the inference performance but also greatly reduces the processing time. To demonstrate the effectiveness of the proposed HFDS structure, the diagnosis system has been tested on a typical secondary transmission system of Tainan area in the Taipower System. The test data contain 360 test cases, including 220 single-fault, 100 double-fault, and 40 triple-fault cases. Composition of test cases and correct rate of the identification is displayed in Table 1. Results of the test show that the proposed HFDS can obtain an accuracy of fault section estimation as high as 100 percent, and in most of the test cases (93.52 percent in average) the HFDS can identify the fault type to provide operators further information on fault situation. In some of the test cases (6.48 percent), due to lack of more information, e.g., only alarm signals of pilot relays offered, the HFDS can just provide correct fault section estimation but fails to identify the fault type. In the above test cases, the processing time, from the time-stamped alarm signals arriving at final steady state to the displaying of the fault situations in the screen of the console, took an average of only 21.63 ms. Rapid and accurate diagnosis results make the proposed HFDS feasible in on-line and real-time application for the practical Taipower system. The work reported in this paper will be incorporated with the Restoration Support Subsystem and the Fault Location Subsystem to form a complete substation-level computer aided decision support system. Owing to the flexibility, inheritance, and reusable coding features of OOP scheme, the proposed system can be expected to be easily integrated into the more and more complicated and increasing functions of energy management system in the future.
|Number of pages||2|
|Journal||IEEE Power Engineering Review|
|Publication status||Published - 1997 Dec 1|
All Science Journal Classification (ASJC) codes
- Electrical and Electronic Engineering