B2RL

B2RL数据集是由加州大学圣地亚哥分校和亚马逊合作开发的，专注于建筑批量强化学习（BRL）领域。该数据集收集了来自实际建筑管理系统的数据以及模拟环境中生成的缓冲区，包含约260条数据记录。创建过程中，数据被处理为马尔可夫决策过程（MDP）元组，以支持BRL模型的基准测试。B2RL数据集的应用领域主要集中在建筑控制优化，旨在通过强化学习技术提高建筑能效和居住舒适度。

The B2RL dataset was co-developed by the University of California, San Diego and Amazon, focusing on the field of building batch reinforcement learning (BRL). This dataset collects data from real building management systems and buffers generated in simulated environments, containing approximately 260 data records. During its creation, the data was processed into Markov Decision Process (MDP) tuples to support benchmarking of BRL models. The application scenarios of the B2RL dataset mainly focus on building control optimization, aiming to improve building energy efficiency and occupant comfort through reinforcement learning techniques.

搜集汇总

背景与挑战

背景概述

B2RL是一个开源建筑批量强化学习数据集，旨在为离线强化学习提供基准数据。它包含真实建筑传感器数据和模拟环境生成的缓冲区，真实数据基于学校建筑一年读数，模拟数据使用Sinergym框架生成多种策略缓冲区。数据集定义了状态、动作和奖励变量，用于支持建筑HVAC控制等应用，是首个开源建筑批量强化学习数据集。