簡化的蟻群最佳演算法與其在模糊類神經網路之應用

Abstract

在本論文中，提出一個簡化的蟻群最佳演算法與其在模糊類神經網路之應用。傳統上螞蟻群聚最佳演算法屬於在解離散組合最佳化問題，其需要複雜的演算流程。因此，在本篇論文中提出一個連續最佳化的方法。並將此方法與模糊類神經網路做結合，應用於函數近似、非線性系統的模組化、以及非線性系統的控制。針對函數近似與非線性系統的模組化的應用，模糊類神經網路的權重值因子能透過離線學習的程序來做調整。於非線性系統控制應用上，分別考慮多輸入多輸出、狀態與輸出回授之非線性系統，藉由即時調整模糊類神經網路的參數以完成控制目的。在多輸入多輸出非線性系統控制設計中，其控制觀點結合了倒階的設計技術與模糊類神經網路。根據其控制的技術，其模糊類神經倒階控制器經由蟻群最佳演算法的方法來做參數的即時調整。針對狀態或輸出回授控制設計，藉著使用直接型控制器的設計概念，與在本篇論文中所提出的簡化的蟻群最佳演算法為基礎的B-spline模糊類神經控制器來控制非線性系統。為了要線上調整這些參數與評估閉迴路系統穩定性的目的，我們提出一個能量適應函數於簡化的蟻群最佳演算法中。並藉著Lyapunov函數來做閉迴路系統的穩定度分析。另外，為了保證閉迴路系統的穩定度，監督式控制器會被利用來與簡化的蟻群最佳演算法為基礎的B-spline模糊類神經控制做結合。最後從模擬結果驗證其所提出方法的可行性與適用性。

In this thesis, a compact ant colony optimization algorithm (CACOA) of fuzzy-neural networks is proposed. Traditionally, ant colony optimization algorithms solve discrete combinatorial optimization problems, and always need complicated operation procedures. Therefore, a continuous ant colony optimization algorithm is proposed for function approximation, nonlinear system modeling, and nonlinear system control. For function approximation and nonlinear system modeling, the weighting factors of the fuzzy-neural networks can be tuned through off-line learning procedure. For a class of multiple-input multiple-output (MIMO) nonlinear systems, the control scheme incorporates backstepping technique with the fuzzy neural networks, and the adjusted parameters of the fuzzy neural networks are tuned on-line via the CACOA approach. For state-feedback and output-feedback control, based on the direct adaptive control approach, a B-spline fuzzy-neural controller using CACOA is proposed to control a class of nonlinear systems. For the purpose of on-line tuning these parameters and evaluating the stability of the closed-loop system, an energy fitness function is included in the CACOA approach. The stability of the closed-loop system is analyzed by means of Lyapunov functions. In addition, in order to guarantee the stability of the closed-loop nonlinear system, a supervisory controller is incorporated into the CACOA-based B-spline fuzzy neural controller. Finally, the simulation results demonstrate the feasibility and applicability of the proposed method.