針對目前火災探測技術難以滿足實際需要的問題，在分析RBF 網絡結構特點及最近鄰聚類學習算法的基礎上，提出用RBF 神經網絡建立火災探測器模型，以火災初期實驗得到的環境溫度、煙霧濃度、CO 含量為輸入，以明火概率、陰燃火概率、無火概率為輸出對RBF 網絡進行訓練，并進行仿真試驗，結果表明，實際輸出與期望輸出的相差較小。
關鍵詞：徑向基函數（RBF）；神經網絡；最近鄰聚類算法；火災探測器模型
Abstract: In view of the problem that presently, the fire detective technology difficultly meet the actual needs, and on the basis of analyzing the characteristic of RBF network structure and the Nearest Neighbor-Clustering Algorithm, there comes a proposal that to use the RBF neural network to establish fire detector model .Taking the ambient temperature, the smog density, the CO content which obtains by the experiment of fire initial period as the input, and flame probability, glowing fire probability and misfire probability as the output. Researchers carry on the training to the RBF network. Meanwhile, they carry on the simulation experiment. As a result, there is little difference between the actual output and expected output.
Key word: Radial Basis Function (RBF), Neural networks ，Nearest Neighbor-Clustering
Algorithm, Fire detector model