PSO、AOA與MINLP演算法應用於 V2G/G2V即時充放電規劃中的性能比較與分析研究

Abstract

在全球淨零排放的趨勢下，電動車已逐漸成為電網中重要的一環。透過車輛對電網(Vehicle-to-Grid, V2G)及電網對車輛(Grid-to-Vehicle, G2V)技術，電動車不再僅是單向的電力負載，而是轉變為分散式儲能單元。V2G技術能藉由輔助服務提供電網靈活性、可靠性及穩定性；然而，頻繁的充放電過程亦加速了電動車電池的老化與損耗。在本研究中，我們將使用者總成本，包含能源交易淨成本與電池退化成本之最小化設定為目標，並在即時排程的框架下，探討不同演算法在求解此一複雜最佳化問題上的性能。 為解決此一非線性、多維度且具備即時運算需求的最佳化問題，本研究比較了三種不同性質的演算法。除了做為精確解標竿的混合整數非線性規(Mixed-Integer Nonlinear Programming, MINLP)外，亦探討了粒子群演算法(Particle Swarm Optimization, PSO)與阿基米德浮力演算法(Archimedes Optimization Algorithm, AOA)兩種啟發式演算法。考量到原始啟發式演算法在尋找全域最佳解的過程中，可能面臨效率低落或早熟收斂(Premature Convergence)的問題，本研究進一步針對 PSO與AOA進行改良，分別設計出退火型PSO與結合維度學習的改良式AOA (Modified AOA, MDAOA)，以增強其在V2G即時排程問題上的求解效能。

Amid the global trend toward net-zero emissions, electric vehicles (EVs) have evolved from unidirectional electrical loads into distributed energy storage units through Vehicle-to-Grid (V2G) and Grid-to-Vehicle (G2V) technologies. V2G services enhance grid flexibility, reliability, and stability ; however, frequent charging and discharging cycles accelerate the aging and degradation of EV batteries. This thesis aims to minimize the total user cost, comprising the net cost of energy transactions and battery degradation expenses. Specifically, it investigates the performance of various algorithms in solving this complex optimization problem within a real-time scheduling framework. To address this non-linear, multi-dimensional optimization problem characterized by real-time computational constraints, this study evaluates three distinct algorithms. In addition to Mixed-Integer Nonlinear Programming (MINLP), which serves as a benchmark for the exact solution, two heuristic algorithms are investigated: Particle Swarm Optimization (PSO) and the Archimedes Optimization Algorithm (AOA). Recognizing that standard heuristic algorithms may suffer from inefficiency or premature convergence, this research proposes enhancements to both methods. Specifically, an Annealing-based PSO and a Modified AOA with Dimension Learning (MDAOA) are developed to improve performance in solving the real-time V2G scheduling problem.