The paper describes a GIS approach to identify the most suitable areas for growing the Paulownia tree in the Province of Bergamo;this work will become the subject of a dissertation thesis at the School of Engineering ...The paper describes a GIS approach to identify the most suitable areas for growing the Paulownia tree in the Province of Bergamo;this work will become the subject of a dissertation thesis at the School of Engineering of the University of Bergamo. Paulownia is quite an ancient tree, well known especially in the far east countries, for its wonderful flowers and the quality of the wood which is a source of certain profit. The paper is focused on the employment of the Paulownia tree against air pollution, since its big leaves absorb high amounts of CO2 in the atmosphere and can capture the small particles of pollutants in the air ten times more than any other tree species. First, the analysis identifies the most favourable areas for its growth in the Province of Bergamo. Secondly, among the latter, the analysis detects the strategic areas for its growth near the most polluted zones. Finally, a quantitative evaluation is carried out about the trees contribution against pollutants in the atmosphere.展开更多
蒙特卡洛树搜索(Monte Carlo tree search, MCTS)将强化学习的反馈优化与生长树的动态规划相结合,在输出当前状态的最佳动作的同时极大地减少了计算量,因此成为开放环境下众多领域智能系统的关键通用方法.但由于计算资源匮乏或者计算成...蒙特卡洛树搜索(Monte Carlo tree search, MCTS)将强化学习的反馈优化与生长树的动态规划相结合,在输出当前状态的最佳动作的同时极大地减少了计算量,因此成为开放环境下众多领域智能系统的关键通用方法.但由于计算资源匮乏或者计算成本昂贵等原因,完全充分地对树结构进行搜索是难以实现的,因此在有限的预算下高效合理地分配计算资源从而获得当前状态下的最优动作是目前研究的一个重要问题.现有大多数算法仅以识别准确率作为性能指标,通过实验对比验证算法性能,缺少对算法的识别误差和影响因素的分析,从而降低了算法的可信性和可解释性.针对该问题,选择基础核心的2名玩家、完全信息、零和博弈场景,提出了固定预算设定下MCTS抽象模型的最优行动识别算法DLU——基于相对熵置信区间的纯探索(relative entropy confidence interval based pure exploration).首先提出了基于相对熵置信区间的估值方法对叶子节点胜率进行估计,其可以从底层提高树节点估值准确性;其次给出了第1层节点值估计、最优节点选择策略以形成完整算法流程;然后推导了DLU算法的识别误差上界,并分析了算法性能的影响因素;最后在人造树模型和井字棋2种场景下验证算法性能.实验结果表明,在人造树模型上基于相对熵的算法类具有更高的准确度,且模型越复杂识别难度越高时,该算法类的性能优势越显著.在井字棋场景下,DLU算法能有效地识别最优动作.展开更多
文摘The paper describes a GIS approach to identify the most suitable areas for growing the Paulownia tree in the Province of Bergamo;this work will become the subject of a dissertation thesis at the School of Engineering of the University of Bergamo. Paulownia is quite an ancient tree, well known especially in the far east countries, for its wonderful flowers and the quality of the wood which is a source of certain profit. The paper is focused on the employment of the Paulownia tree against air pollution, since its big leaves absorb high amounts of CO2 in the atmosphere and can capture the small particles of pollutants in the air ten times more than any other tree species. First, the analysis identifies the most favourable areas for its growth in the Province of Bergamo. Secondly, among the latter, the analysis detects the strategic areas for its growth near the most polluted zones. Finally, a quantitative evaluation is carried out about the trees contribution against pollutants in the atmosphere.
文摘蒙特卡洛树搜索(Monte Carlo tree search, MCTS)将强化学习的反馈优化与生长树的动态规划相结合,在输出当前状态的最佳动作的同时极大地减少了计算量,因此成为开放环境下众多领域智能系统的关键通用方法.但由于计算资源匮乏或者计算成本昂贵等原因,完全充分地对树结构进行搜索是难以实现的,因此在有限的预算下高效合理地分配计算资源从而获得当前状态下的最优动作是目前研究的一个重要问题.现有大多数算法仅以识别准确率作为性能指标,通过实验对比验证算法性能,缺少对算法的识别误差和影响因素的分析,从而降低了算法的可信性和可解释性.针对该问题,选择基础核心的2名玩家、完全信息、零和博弈场景,提出了固定预算设定下MCTS抽象模型的最优行动识别算法DLU——基于相对熵置信区间的纯探索(relative entropy confidence interval based pure exploration).首先提出了基于相对熵置信区间的估值方法对叶子节点胜率进行估计,其可以从底层提高树节点估值准确性;其次给出了第1层节点值估计、最优节点选择策略以形成完整算法流程;然后推导了DLU算法的识别误差上界,并分析了算法性能的影响因素;最后在人造树模型和井字棋2种场景下验证算法性能.实验结果表明,在人造树模型上基于相对熵的算法类具有更高的准确度,且模型越复杂识别难度越高时,该算法类的性能优势越显著.在井字棋场景下,DLU算法能有效地识别最优动作.
