本論文中提出了一個整合型的學徒式學習器，整合分析式與經驗式學習方
式，學習在核能電廠蒸汽產生器管路破裂事故時，避免因動力釋放閥打開
產生的輻射外洩處理知識。本系統均勻的從電廠操作員或模擬器所餵給的
正、負例子中粹取有用的資訊，進行多重錯誤的領域理論修正。本系統採
用漸進式學習方式，但只記錄有用的資訊，可降低儲存全部例子的大量空
間需求；加強負例子的處理能力，可加速領域理論排除負例子的能力，加
速歸納的方向，可節省搜尋時間。

This thesis proposes an integrated learner combining analytic
and empirical learning methods to learn the handling knowledge
that can prevent Power Operated Relief Valves to open causing
radiation release during the Steam Generator Tube Rupture
accident.This learner extracts useful information from both
positive and negative examples, fed by the plant's operators or
the on-line simulator, to deal with multiple errors in the
domain theory. This approach features its incremental learning
technique without rememberingall examples; a stronger negative
examples processing capability to help thedomain theory to fast
recognize future system error; and a domain model , bythe help
of analyzing the example, to help reduce induction space.