【摘要】提出了一種非線性系統(tǒng)的模型辨識(shí)方法。在只有被辨識(shí)系統(tǒng)的輸入輸出數(shù)據(jù)的情況下，利用一種無監(jiān)督的聚類算法來進(jìn)行結(jié)構(gòu)辨識(shí)，從而自動(dòng)獲得模糊規(guī)則庫(kù)，并可以得到模糊系統(tǒng)的初始參數(shù)。在聚類的基礎(chǔ)上，構(gòu)造一個(gè)與之相匹配的模糊神經(jīng)網(wǎng)絡(luò)，用它的學(xué)習(xí)算法來訓(xùn)練網(wǎng)絡(luò)得到一個(gè)精確的模糊模型，從而實(shí)現(xiàn)參數(shù)辨識(shí)。同時(shí)，證明了所構(gòu)造的模糊神經(jīng)網(wǎng)絡(luò)具有通用逼近能力，這個(gè)能力在模糊建模和模糊控制方面非常有用。通過對(duì)兩個(gè)非線性系統(tǒng)辨識(shí)的仿真結(jié)果驗(yàn)證了該方法的有效性。
關(guān) 鍵 詞 模糊神經(jīng)網(wǎng)絡(luò); 結(jié)構(gòu)辨識(shí); 參數(shù)辨識(shí); 系統(tǒng)辨識(shí)

In the last few years, based on neural network and fuzzy system, a lot of new system identification and control methods of nonlinear system have been proposed [1～5]. Despite the fact that these methods are effective in some application area, most of them are only used in parameter identification not in structure identification. The proposed model identification of nonlinear system is composed of two parts: structure identification and parameter identification. At the same time, this FNN has universal approximation capability, a property very useful in, e.g., modeling and control applications.