Technology learning for fuel cells: An assessment of past and potential cost reductions

Fuel cells have gained considerable interest as a means to efficiently convert the energy stored in gases like hydrogen and methane into electricity. Further developing fuel cells in order to reach cost, safety and reliability levels at which their widespread use becomes feasible is an essential prerequisite for the potential establishment of a ‘hydrogen economy’. A major factor currently obviating the extensive use of fuel cells is their relatively high costs. At present we estimate these at about 1100 €(2005)/kW for an 80 kW fuel cell system but notice that specific costs vary markedly with fuel cell system power capacity. We analyze past fuel cell cost reductions for both individual manufacturers and the global market. We determine learning curves, with fairly high uncertainty ranges, for three different types of fuel cell technology – AFC, PAFC and PEMFC – each manufactured by a different producer. For PEMFC technology we also calculate a global learning curve, characterised by a learning rate of 21% with an error margin of 4%. Given their respective uncertainties, this global learning rate value is in agreement with those we find for different manufacturers. In contrast to some other new energy technologies, R&D still plays a major role in today’s fuel cell improvement process and hence probably explains a substantial part of our observed cost reductions. The remaining share of these cost reductions derives from learning-by-doing proper. Since learning-by-doing usually involves a learning rate of typically 20%, the residual value for pure learning we find for fuel cells is relatively low. In an ideal scenario for fuel cell technology we estimate a bottom-line for specific (80 kW system) manufacturing costs of 95 €(2005)/kW. Although learning curves observed in the past constitute no guarantee for sustained cost reductions in the future, when we assume global total learning at the pace calculated here as the only cost reduction mechanism, this ultimate cost figure is reached after a large-scale deployment about 10 times doubled with respect to the cumulative installed fuel cell capacity to date.     

Societal penetration of hydrogen into the future energy system: Impacts of policy,technology and carbon targets

Decarbonization of the energy system is a key goal of the Paris Agreements, in order to limit temperature rises to under 2° Celsius. Hydrogen has the potential to play a key role through its versatile production methods, end uses and as a storage medium for renewable energy, engendering the future low-carbon energy system. This research uses a global model cognizant of energy policy, technology learning curves and international carbon reduction targets to optimize the future energy system in terms of cost and carbon emissions to the year 2050. Exploring combinations of four exploratory scenarios incorporating hydrogen city gas blend levels, nuclear restrictions, regional emission reduction obligations and carbon capture and storage deployment timelines, it was identified that hydrogen has the potential to supply approximately two percent of global energy needs by 2050. Irrespective of the quantity of hydrogen produced, the transport sector and passenger fuel cell vehicles are consistently a preferential end use for future hydrogen across regions and modeled scenarios. In addition to the potential contribution of hydrogen, a shift toward renewable energy and a significant role for carbon capture and storage is identified to underpin carbon target achievement by 2050.     

A novel seasonal decomposition based least squares support vector regression ensemble learning approach for hydropower consumption forecasting in China

Due to the distinct seasonal characteristics of hydropower, this study tries to propose a seasonal decomposition (SD) based least squares support vector regression (LSSVR) ensemble learning model for Chinese hydropower consumption forecasting. In the formulation of ensemble learning model, the original hydropower consumption series are first decomposed into trend cycle, seasonal factor and irregular component. Then the LSSVR with the radial basis function (RBF) kernel is used to predict the three different components independently. Finally, these prediction results of the three components are combined with another LSSVR to formulate an ensemble result for the original hydropower consumption series. In terms of error measurements and statistic test on the forecasting performance, the proposed approach outperforms all the other benchmark methods listed in this study in both level accuracy and directional accuracy. Experimental results reveal that the proposed SD-based LSSVR ensemble learning paradigm is a very promising approach for complex time series forecasting with seasonality.     

Early clinical outcomes after small incision lenticule extraction surgery (SMILE)

Purpose

Dry eye is known to impact on clinical outcomes after laser vision correction and the use of a newer ‘all femtosecond laser’ surgical approach may be associated with less impact on the ocular surface post-operatively. The purpose of this study was to evaluate the early clinical outcomes and tear instability after the first small incision lenticule extraction (SMILE) cases undertaken by three surgeons at a single site in the UK.

基于超松弛迭代的标签传播算法*

针对标签传播算法中存在的问题,将超松弛迭代引入标签传播算法,解决标签序列的优化问题,提出基于超松弛迭代的标签传播算法(ORLP).该算法使用正负标签的方式标记已知样本,通过在近邻点间学习分类的方式预测未知样本的标签信息,同时在每次迭代时都能较好地保留初始标记点的标签信息,以指导下一次的标签传递过程.基于超松弛迭代推导ORLP的标签传播公式,同时证明标签序列的收敛性,得到标签序列的收敛解.实验表明,ORLP具有较高的分类准确率和较快的收敛速度.     

基于学习向量量化神经网络的软件可靠性预测

针对传统的软件可靠性预测模型在实际应用中存在预测泛化性能不佳等问题,提出一种基于学习向量量化（LVQ）神经网络的软件可靠性预测模型。首先分析了LVQ神经网络的结构特点以及它与软件可靠性预测的联系,然后运用该网络来进行软件可靠性的预测,并基于美国国家航空航天局（NASA）软件数据项目中的实例数据集,运用Matlab工具进行了仿真实验。通过与传统预测方法的对比,证明该方法具有可行性和较高的预测泛化性能。     

基于文本情感分析的城市公园使用感知评价研究 ——以重庆36个公园为例

公园是城市重要的公共空间,是居民 亲近自然、放松身心的主要场所。用户在访问公 园过程中,多种感知要素影响其体验和评价,通 过分析其影响机制和特征可以针对性地改善使 用体验,提高规划的人文关怀。本文以重庆36 个公园为例,通过网络爬虫技术采集海量用户点 评数据,使用百度深度学习模块进行文本情感分 析,共挖掘10个感知要素。通过分析要素感知频 率、感知倾向以及不同公园类型自身感知特征, 提取影响公园感知的6个关键因素,其中自然景 观、绿化覆盖、公园规模、声音气味4个要素可 以通过提高可视性等措施来增加积极感知,公 园服务设施、地形起伏2个要素是公园规划建设 中应重点改善的内容。本文希望通过使用感知 导向的主观评价,为人性化公园和宜居城市建 设提供指导。     

半监督谱聚类特征向量选择算法

对于一个K类问题,Ng-Jordan-Weiss(NJW)谱聚类算法通常采用数据规范化亲和度矩阵的前K个最大特征值对应的特征向量作为数据的一种表示。然而,对于某些模式识别问题,这K个特征向量不一定能够体现原始数据的结构。文中提出一种半监督谱聚类特征向量选择算法。该算法利用一定量的监督信息寻找能够体现数据结构的特征向量组合,进而获得优于传统谱聚类算法的聚类性能。UCI标准数据集和MNIST手写体数据集上的仿真实验验证该算法的有效性和鲁棒性。     

学习领域课程及实施对策分析

德国学习领域课程以工作过程结构理论为基础,是与建立在传统学科结构理论基础上的学科课程相对的一种课程模式。与学科课程相比,前者更符合职业学校学生的认知特点。根据学习领域课程开发与具体职业相结合这一特征,提出了学习领域课程开发的两个关键问题：一个是从职业活动形态向职业学校课程形态转化,一个是从职业学校课程形态向职业活动形态转化。认为课程开发的质量取决于对这两个关键问题的把握。要推动优秀课程的开发,就要从提高影响转化的两个重要因素——教师和实训设施入手。     

发动机机油泵供油特性的神经网络建模

结合试验研究结果,通过对影响机油泵供油特性影响因素的分析,建立了用于分析机油泵供油特性的BP神经网络仿真模型,并利用该模型对机油泵的温度特性、转速特性和压力特性进行了预测分析,取得良好的仿真效果。通过对不同样本训练得到的神经网络模型仿真结果和试验结果的比较分析,验证了采用正交法设计该网络模型学习样本的可行性,指出利用正交法设计学习样本时,只要位级选择适当,就可以大大减少学习样本的数量,并取得满意的预测效果。