粒子群演算法應用於氫能水下載具之能源管理系統最佳化

Abstract

本研究以英國國家海洋中心（National Oceanography Centre, NOC）所開發之水下載具-Autosub-6000為例，透過MATLAB/Simulink建立氫能水下載具之縱向動態及相關次系統模型，透過此系統進行縱向動態模擬，觀察水下載具推力、螺槳轉速及速度變化，另外也針對能量分配策略進行設計，利用if-else邏輯設計基本規則庫控制（Rule Based Control, RBC）針對燃料電池與鋰離子電池進行能量分配，此策略較難找尋最佳能量分配比例，故本研究開發粒子群演算法（Particle Swarm Optimization, PSO），比較兩種策略的能耗，相較於RBC策略PSO可改善約25%的能耗，為了確認PSO已達最佳能耗表現，本研究透過全域搜索（Global Search Algorithm, GSA），找出目標函數的最佳解，與PSO進行比較，分析兩者能耗表現之均方根誤差（Root Mean Square Error, RMSE），其RMSE約為0.37kWh，表示PSO已收斂至最佳解，在殘電量（State of Charge, SOC）的部分，RMSE僅為1.99%，顯示兩者有高度相似性。另外在MATLAB/Simulink進行模型迴路測試（Modelling In the Loop, MIL）之結果不一定能用在真實情境，故本研究建置硬體迴路測試系統（Hardware In the Loop, HIL），進行即時模擬（Real-Time），分析MIL及HIL之能耗及SOC，RMSE結果顯示，能耗表現的誤差約為2.05kWh，SOC變化誤差僅有1.3%。

This research utilizes the Autosub-6000 underwater vehicle, developed by the UK's National Oceanography Centre (NOC), as a case study. We've established a longitudinal dynamic model and related subsystem models for a hydrogen fuel cell underwater vehicle using MATLAB/Simulink. Through this system, we conducted longitudinal dynamic simulations to observe changes in the vehicle's thrust, propeller speed, and velocity during operation. Furthermore, we designed an energy distribution strategy. Initially, a basic Rule-Based Control (RBC) system was designed using if-else logicfor energy distribution between the fuel cell and lithium battery. Recognizing the difficulty in finding an optimal distribution ratio with RBC, this study developed a Particle Swarm Optimization (PSO) algorithm. By comparing the energy consumption of these two strategies, PSO demonstrated an improvement of approximately 25% in energy consumption compared to the RBC strategy. To confirm PSO's optimal energy performance, we employed the Global Search Algorithm (GSA) to identify the optimal solution for the objective function and compared it with PSO. The Root Mean Square Error (RMSE) for their energy consumption was approximately 0.37 kWh, indicating that PSO had converged to the optimal solution. Regarding the State of Charge (SOC), the RMSE was only 1.99%, showing a high degree of similarity between the two strategies. Since Model-in-the-Loop (MIL) test results in MATLAB/Simulinkmay not always translate directly to real-world scenarios, this research established a Hardware-in-the-Loop (HIL) testing system for real-time simulation. We analyzed the energy consumption and SOC changes in both MIL and HIL. The RMSE analysis revealed an energy consumption error of approximately 2.05 kWh and an SOC change error of only 1.3%, demonstrating a high degree of similarity between MIL and HIL results.