关键词： Building energy management   Data predictive control   Model predictive control   Physics-informed Machine Learning   Validation in experiment  

摘要： Because physics-based building models are difficult to obtain as each building is individual, there is an increasing interest in generating models suitable for building MPC directly from measurement data. Machine learning methods have been widely applied to this problem and validated mostly in simulation;there are, however, few studies on a direct comparison of different models or validation in real buildings to be found in the literature. Methods that are indeed validated in application often lead to computationally complex non-convex optimization problems. Here we compare physics-informed Autoregressive-Moving-Average with Exogenous Inputs (ARMAX) models to Machine Learning models based on Random Forests and Input Convex Neural Networks and the resulting convex MPC schemes in experiments on a practical building application with the goal of minimizing energy consumption while maintaining occupant comfort, and in a numerical case study. The building has a water-based emission system and is located in temperate climate. We demonstrate that Predictive Control leads to savings between 26% and 49% of heating and cooling energy, compared to the building's baseline hysteresis controller. Moreover, we show that all model types lead to satisfactory control performance in terms of constraint satisfaction and energy reduction. However, we also see that the physics-informed ARMAX models have a lower computational burden, and a superior sample efficiency compared to the Machine Learning based models. Moreover, even if abundant training data is available, the ARMAX models have a significantly lower prediction error than the Machine Learning models, which indicates that the encoded physics-based prior of the former cannot independently be found by the latter.

关键词： Machine learning  

摘要： Ranunculaceae is a botanical source for various pharmaceutically active compounds, which has been commonly utilized in traditional Chinese medicine. Increasing interest in Ranunculaceae pharmaceutical resources has led to a taxonomical study of this family, which might provide new insight to understand its diversification, relationship and phylogenetic position, and further to find new medicinal resources and promising compounds. In this study, we used the machine learning method to explore the classification of the medicinal Ranunculaceae family. 204 species representing 17 genera of the Ranunculaceae family were collected from the TCMID with their 1280 active compounds composed of structure-based fingerprints. After the construction of species-compound and genus-compound matrices, CNNs and Ext fingerprints were determined as the best machine learning method and fingerprint type using ACC and F-score as clustering criteria, respectively. We found that taxonomical classification within the Ranunculaceae family could be accurately predicted, especially at the genus level with a top ACC of 0.86 and an F-score of 0.85. The top features of compounds that were important for the classification of 17 genera were also identified, and thus some genera with high medicinal values were associated with characteristic cis and (or) trans features. As far as we know, this is the first time that some genera are found to be associated with the structural features of compounds.

关键词： 机器学习   随机森林   磁性基态   磁矩  

摘要： 磁性材料是信息时代重要的基础材料,不同的磁性基态是磁性材料广泛应用的前提,其中铁磁基态是高性能磁性材料的关键要求.本文针对材料项目数据库中的无机磁性材料数据,采用机器学习技术实现无机磁性材料铁磁、反铁磁、亚铁磁和顺磁基态的分类以及无机铁磁性材料磁矩的预测.提取了材料的元素和结构属性特征,通过两步式特征选择方法分别为磁性基态分类和磁矩预测筛选了20个材料特征,发现材料特征中的电负性、原子磁矩和原子外围轨道未充满电子数对两种磁性性能具有重要贡献.基于机器学习的随机森林算法,构建了磁性基态分类模型和磁矩预测模型,采用10折交叉验证的方法对模型进行定量评估,结果表明所构建的模型具有足够的精度和泛化能力.在测试检验中,磁性基态分类模型的准确率为85.23%,精确率为85.18%,召回率为85.04%,F1分数为85.24%;磁矩预测模型的拟合优度为91.58%,平均绝对误差为0.098μB/atom.本研究为无机铁磁性材料的高通量分类筛选与磁矩预测提供了新的方法和选择,可为新型无机磁性材料的设计研发提供参考.

关键词： 虚拟化技术   云计算   负载预测   数据中心   大数据   机器学习方法   数理统计   计算处理  

摘要： 随着大数据的发展,需要加强对数据的计算处理工作。云计算能够根据用户需求进行数据中心资源的分配工作,能够显著的提升资源效率和网络灵活性,节约企业的经济成本,具备较好的经济效益。要明确主机负载的变化特征,并根据不同的应用情况选择最佳的负载预测算法,历史数据的分析和采集可以采用不同的方式,定量预测是综合数理统计、概率论来进行分析,根据使用的机器学习方法来完成时间序列数据的预测,具备非常强的现实工业价值。

关键词： 高维表示方法   支持向量机   离心泵   扭曲叶片   动能方程  

摘要： 提出将高维表示方法与机器学习中的支持向量机方法结合对离心泵扭曲叶片优化。选用1台中比转速离心泵为研究对象,通过对扭曲叶片3条叶片型线的参数化,分离控制变量并确定代理模型的训练空间。经过对离心泵扭曲叶片水力模型的算法学习,得到以叶片型线参数为自变量,效率为目标函数的离心泵代理模型,运用遗传算法求解并反馈叶片型线参数。采用数值模拟及试验的方法验证了代理模型的预测结果,并从动能方程的角度分析了优化前后扭曲叶片叶轮内流场的变化。结果表明:在设计工况点,优化后的扭曲叶片离心泵数值模拟效率值比原型泵提高了1.72%,扬程提升了0.41 m,试验效率值比原型泵提高了1.5%,扬程提升了0.35 m。

关键词： 粒子群-神经网络   随机森林算法   经验正交函数   气候预测   汛期降水  

摘要： 使用1991-2021年7-9月广西90个地面气象观测站降水量、NCEP/NCAR月再分析资料和国家气候中心BCC_CSM1.1气候模式回报资料,研究建立基于粒子群-神经网络、随机森林算法的广西后汛期降水气候预测模型,并对2016年-2021年预测进行应用试验。结果表明,基于粒子群-神经网络、随机森林算法的后汛期降水预测Ps得分比逐步回归方法分别提高了2.78分、2.5分,比气候模式分别提高了29.22分、28.94分,预测能力有明显的提升。

关键词： 废水排放量   社会经济决定因素   趋势预测   机器学习方法   任务相关性  

摘要： 中国的水污染是一个非常严重的问题,为了控制水污染,对中国未来废水排放量进行准确预测具有十分重要的意义。从影响废水排放量的潜在变量中筛选出GDP、人均GDP和产业结构三个关键预测变量,采用1998—2017年中国省域数据和基于任务相关性的机器学习方法对各省份废水排放量趋势进行预测。实验结果表明,所建立的预测模型能够保持线性回归模型的可解释性,并获得较高的预测精度(RMSE、MAE和MAPE分别为0.0882、0.0643和0.5533)。预计全国年废水排放总量2025年最低将达到9081209万吨,最高9903864万吨;2030年最低将达到10864311万吨,最高13081922万吨。政府部门可根据预测结果制定废水减排计划和经济发展规划,在发展经济的同时维持生态环境的平衡。

关键词： Biome reconstruction   Random forest algorithm   Biomisation method   Pollen data   Last Glacial Maximum   Tibetan Plateau  

摘要： Historical biome changes on the Tibetan Plateau provide important information that improves our understanding of the alpine vegetation responses to climate ***,a comprehensively quantitative reconstruction of the historical Tibetan Plateau biomes is not possible due to the lack of quantitative methods that enable appropriate classification of alpine biomes based on proxy data such as fossil pollen *** this study,a pollen-based biome classification model was developed by applying a random forest algorithm(a supervised machine learning method)based on modern pollen assemblages on and around the Tibetan Plateau,and its robustness was assessed by comparing its results with the predictions of the biomisation *** results indicated that modern biome distributions reconstructed using the random forest model based on modern pollen data generally concurred with the observed zonal *** random forest model had a significantly higher accuracy than the biomisation method,indicating the former is a more suitable tool for reconstructing alpine biome changes on the Tibetan *** random forest model was then applied to reconstruct the Tibetan Plateau biome changes from 22 ka BP to the present based on 51 fossil pollen *** reconstructed biome distribution changes on the Tibetan Plateau generally corresponded to global climate changes and Asian monsoon *** the Last Glacial Maximum,the Tibetan Plateau was mainly desert with subtropical forests distributed in the *** the last deglaciation,the alpine steppe began expanding and gradually became zonal vegetation in the central and eastern *** meadow occupied the eastern and southeastern areas of the Tibetan Plateau since the early Holocene,and the forest-meadow-steppe-desert pattern running southeast to northwest on the Tibetan Plateau was established *** the mid-Holocene,subtropical forests extended north,which reflected the“optimum”***

关键词： 数据中心   PUE指标   能效优化   机器学习   神经网络   灵敏度分析  

摘要： 本文研究如何利用机器学习方法来降低数据中心电能使用效率(power usage effectiveness,PUE)指标.现阶段以Google公司为代表的工业界使用的神经网络模型考虑的特征数量较少,且仅考虑单一特征变化对PUE指标的影响,缺少对特征之间耦合特性的分析.此外,机器学习方法对数据集的质量和数量要求都很高,并且落地实施过程和结论判断皆容易受到噪声干扰,整体难度较大.现阶段学术界和工业界尚缺少对具体优化案例的详细阐述.本文对现阶段利用神经网络模型优化PUE指标的方法进行改进,增加了特征维数,提高了预测精度,取得了超出Google公司PUE模型预测精度的结果.利用历史样本轨道,使用统计方法近似得出特征之间的耦合特性,并代入灵敏度分析中,得到更加精确的分析结果.提出基于灵敏度分析的冷却系统参数设计优化方案;利用腾讯华北某数据中心的海量数据和现场条件,实施制冷系统参数设定优化的实验,实验效果证明了优化方案的有效性.

关键词： 机器学习   粒度剖面预测   测井曲线   地质纵向连续性  

摘要： 地层砂的粒度特征值d50(筛析曲线累计质量分数50%对应的粒径值,μm)是防砂设计中的关键参数,为获得粒度纵向分布剖面,开展了基于机器学习方法的储层粒度与测井曲线响应关系研究。经典机器学习往往缺少模型内部的特征提取过程,而且采用单一采样点作为输入,缺失相邻数据关联关系反映层位信息。考虑到储层的地质连续性,利用测井曲线趋势和背景信息,将深度相邻数据点作为机器学习特征值,提出了一种基于多采样点的粒度剖面预测方法,构造和训练了基于随机森林(Random Forest)、支持向量机(Support Vector Machine)、Xtreme Gradient Boosting Tree(XGBoost)、人工神经网络(Artificial Neural Network)的预测模型。研究结果表明,与单点映射模型相比,考虑储层纵向地质连续性的各模型预测精度均高于单点预测,其中五点映射的ANN模型(ANN-5)预测效果最好,测试集d50预测相关系数最高为0.819,误差MAE最小为9.59,证实了多个采样点作为输入隐含利用了部分地层信息,有效地提高了预测精度。研究了特征点密度对模型准确率的影响,对训练集二维输入空间中样本的特征点高斯核密度分布以及测试集样本点处的训练集特征点密度进行估算,得出在高密度区域中的测试集样本点的RMSE普遍较低。当增加训练样本数量时,模型预测精度将进一步提高。采用层次分析法确定影响模型选择各因素的权重,通过模糊综合评判法优选机器学习模型,根据优选出的模型对临近区块储层粒度剖面进行预测,预测结果很好地捕捉了粒度变化趋势,模拟了其峰值。