智慧型神經演算法之自走車路徑控制

Fuzzy Gaussian Neural Network Applied to Robot Car Path Control

蒙以正(Yi-Jen Mon) ； 林家葳(Jia-Wei Lin)

《南亞學報》 29期 (2009/12) Pp. 347-363

https://doi.org/10.6989/JN.200912.0347

摘要

本文探討以智慧型類神經演算法發展控制法則，進行三階自走車多輸入多輸出（multi-inputs multi-outputs:MIMO）模型之模擬研究。路徑分為圓形與方型路徑，並以傳統的PID控制方法與之比較。一般對於單輸入單輸出（single input single output:SISO）的系統，以PID（Proportional integration derivative）控制其實效果都不錯，但對於MIMO系統，其實設計的方法很多，例如LQR（Linear quadratic regulator）或LMI（Linear matrix inequality）或Pole placement等均屬之。然而這些方法必須有明確的模型發展，當用在實際系統時，由於有非確定性（uncertainty）存在，可能導致設計過當或設計不足等缺失。智慧型演算法可以進行記憶學習，並將控制法則建立在記憶庫，可在設計階段不考慮太多非確定性，使設計的結果滿足環境變化需求。本文模擬結果顯示，以Fuzzy Gaussian Neural Network（FGNN）之智慧型類神經演算法控制之結果能滿足設計要求並優於傳統PID設計方法。

關鍵字

Fuzzy Gaussian Neural Network ； 智慧型類神經演算法 ； 自走車控制

並列摘要

This paper uses intelligent algorithm to develop the control rules applied to three order multi-inputs multi-outputs (MIMO) robot car model. The simulation paths have circular and square types, and the simulations are compared with traditional PID control.Generally the PID control is suitable for single input single output (SISO) systems, but for the MIMO systems, there are a lot of methods such as LQR, LMI or Pole placement etc.. However these methods must have an explicit model to develop in the actual system, due to the uncertainty existence, this may cause some shortages such as overload design or underload design.The intelligent algorithm has ability of remembering and learning. Meanwhile, it has also the ability to build up control rules in the memory storage. In designing stage, it has good abitility to cope with the environment uncertainties. The simulation results demonstrate that the Fuzzy Gaussian Neural Network control not only can satisfy design requirements but also better than a traditional PID design method.