船舶於海上航行時，其運動行為經常伴隨著海況的變化而改變，這些外在環境變動產生之影響通常是隨機且不易預測的，很難建立其精確的數學模式。因環境干擾而產生之模式不穩定性，常使得傳統的控制法則，即依據受控體模式來設計之自航器較難獲得良好的控制效果。而類神經網路與模糊系統等控制理論，並不需要受控體精確的數學模式，對於受控體模式參數的變化也有較強的適應能力。本文即在於探討類神經網路與模糊理論之控制法則，針對船舶操縱進行自航器設計，並結合兩者優點，提出以適應性網路模糊推論系統為架構並利用函數關係及直接類神經網路即時調整量化因子等，以增加系統反應速率，並降低超越量。最後利用本文所設計出各類型之非模式需求型自航器，以及傳統式需本船模式設計之內模式自航控制器，分別針對航向保持，以及對於預設參考目標之軌跡追蹤成效進行比較分析。於路徑點軌跡追蹤模擬中，本文結合視線導航(line-of-sight guidance)法計算參考航向角，經由一序列之航向變化控制，完成路徑點(waypoints)軌跡追蹤控制試驗。總結，本研究使用MATLAB所提供之模糊邏輯工具箱(fuzzy logic toolbox)及類神經網路，並結合Simulink進行模擬試驗，完成了航向保持、路徑點軌跡追蹤等模擬，確認了本文所提出之各項自航器設計法之可行性。

The ever changing sea condition makes it difficult to use a single mathematical model in describing the dynamic behavior of a ship. Hence, the performance of traditional model-based autopilot is limited due to the so called model uncertainty. On the contrary, the fuzzy logic and artificial neural network controller design methods do not require
explicit modeling of the plant to be controlled; hence, better robustness and adaptive capability can be expected. This work aims at combining the robustness property associated a fuzzy logic controller and the learning capability of an artificial neural network. An ANFIS (Adaptive Network_based Fuzzy Inference System) framework has been proposed that significantly reduces the efforts required in the selection of appropriate membership functions through proper learning process.
Specifically, a function_based tuning method and a direct neural network have been employed to increase the response speed, while reducing the amount of overshoot. The Matlab/Simulink and related toolboxes are used in the simulation experiments and the line of sight guidance method has been adopted in computing the reference heading needed in the track-keeping mission. A series of course-changing and track-keeping
simulations has been conducted to demonstrate the advantage of the proposed ANFIS design framework over the traditional internal model control (IMC), fuzzy, and neural network design methods.