臺灣博碩士論文加值系統

油中氣體分析技術常被用以檢測油浸式變壓器之潛伏性故障，以確保設備安全與供電品質。若能儘量提升診斷方法之正確率，應具正面意義。本文以主成分分析方式觀察常用之Rogers改良法之結果，發現若欲提升其正確率，可能收效不彰。因此，嘗試以類神經網路進行診斷，並與常用之氣體模式比對法、判別分析法，相互比較，以驗証其可行性。經以台電實際之變壓器故障案例，進行測試，其結果顯示，類神經網路法所具之正確率，應在90%以上，且較氣體模式比對法及判別分析法之所得略高。有見於類神經網路技術之應用潛力，本文所提方法應具深入研究之價值。

The dissolve gas analysis techniques have been frequently adopted to diagnose the incipient fault of oil-filled power transformers such that the quality of energy supply and equipment security can be guaranteed. Therefore, it is meaningful to promote the accuracy of diagnostic methods.To increase the accuracy of conventionally-used Rogers ratio method(RRM) may not be justified from the viewpoint of principal components analysis which is applied on the results of RRM. Based on this observation, a neural network based diagnostic method is proposed. This method is also compared with two conventionally- used methods, gas pattern analysis(GPA) and discriminant analysis(DA), to investigate its feasibility.Practical data of faulted case from Taiwan Power Company have been utilized to test the proposed method. From the simulation results, its accuracy is above 90% which is slightly better than GPA and DA. Owing to the potential of the neural network technique, the proposed method is justified to conduct further study