利用計算智能處理遊戲團隊平衡

Abstract

在本篇論文裡，我們提出了一個新的團隊能力平衡系統 (TABS)，以協助遊戲設計師評量一個角色扮演團隊戰鬥類型的遊戲內，任兩個團隊的能力設定是否平衡。TABS 利用基因演算法或粒子群優化，來訓練人工神經網路控制器。訓練成果最傑出的控制器會被選擇並應用於團隊平衡度的評估。另外，我們也提出了教練訓練法這種訓練模式，它能公平地訓練控制器。為了加速訓練的程序，我們運用了多執行緒技術，讓控制器能在數個獨立的遊戲空間裡平行進行訓練。在個案研究中，我們把TABS 應用到我們自行設計的MagePowerCraft 遊戲來進行驗證與實驗。而實驗結果顯示我們的系統效能頗令人滿意。

In this thesis, we propose a novel Team Ability Balancing System (TABS) to assist game designers to evaluate whether the team ability settings of any two teams are balanced in a role-playing combating game. TABS uses artificial neural network controllers which are trained by either genetic algorithm or particle swarm optimization. The best-trained controllers are chosen and applied for the team balance evaluation. Additionally, we also propose the Train-by-Coaches scheme which is useful for training controllers in a fair manner. In order to speed up the training process, we apply the multi-threading techniques to train the controllers in several independent game spaces in parallel. In a case study, we apply TABS to the in-house designed game, MagePowerCraft, for validations and experiments. Experimental results show that the performance of our system is quite satisfactory.