快速近似计算Shapley值的归因解释方法

Shapley值归因解释方法虽然能更准确量化解释结果, 但过高的计算复杂度严重影响了该方法的实用性. 本文引入KD树重新整理待解释模型的预测数据, 通过在KD树上插入虚节点, 使之满足TreeSHAP算法的使用条件, 在此基础上提出了KDSHAP方法. 该方法解除了TreeSHAP算法仅能解释树结构模型的限制, 将该算法计算Shapley值的高效性放宽到对所有的黑盒模型的解释中, 同时保证了计算准确度. 通过实验对比分析, KDSHAP方法的可靠性, 以及在解释高维输入模型时的适用性.     

珠江口盆地珠一坳陷古近系高自然伽马砂岩形成机制及油气地质意义

运用放射性元素寻找油气是一种非常规油气勘探手段。近年来,在珠江口盆地珠一坳陷富烃凹陷周边古近系钻遇高自然伽马(GR)砂岩,其GR值(100~300 API)甚至大于同区泥岩的GR值(100~200 API)。为了弄清该特殊现象背后的地质意义,对珠江口盆地珠一坳陷古近系高自然伽马砂岩开展了铀(U)、钍(Th)、钾(K)等3种元素含量与GR值的相关趋势线分析,从井震特征、岩性特征及矿物成分特征等入手分析了砂岩GR值增高的主要原因及成因机制,探讨了放射性元素聚集的条件、运移通道、驱动力以及油气意义。结果表明:西江、惠州地区由U含量增高导致砂岩GR值偏高,恩平、番禺地区由K,Th含量增高导致砂岩GR值偏高;砂岩GR值增高有两大成因机制,一是地下流体带来的放射性元素离子U4+在氧化-还原面处富集后导致地层GR值偏高,这种特殊现象说明在具有连通基底大断裂旁的圈闭中,U4+的富集指示了曾经油气的存在,证实了研究区油气运移通道的有效性,对于油气藏的预测有着非常重要的指导性意义,二是地表流体带来的含放射性元素的矿物大量沉积后导致地层GR值偏高,含放射性元素矿物性质不稳定,可指示近源供给的存在,对于判断物源及沉积环境有着非常重要的意义。该研究成果为预测研究区油气成藏有利区带提供了依据。     

Multidirectional focus measure for accurate three-dimensional shape recovery of microscopic objects

Shape from focus (SFF) is a technique to recover the shape of an object from multiple images taken at various focus settings. Most of conventional SFF techniques compute focus value of a pixel by applying one of focus measure operators on neighboring pixels on the same image frame. However, in the optics with limited depth of field, neighboring pixels of an image have different degree of focus for curved objects, thus the computed focus value does not reflect the accurate focus level of the pixel. Ideally, an accurate focus value of a pixel needs to be measured from the neighboring pixels lying on tangential plane of the pixel in image space. In this article, a tangential plane on each pixel location (i, j) in image sensor is searched by selecting one of five candidate planes based on the assumption that the maximum variance of focus values along the optical axis is achieved from the neighborhood lying on tangential plane of the pixel (i, j). Then, a focus measure operator is applied on neighboring pixels lying on the searched plane. The experimental results on both the synthetic and real microscopic objects show the proposed method produces more accurate three-dimensional shape in comparison to conventional SFF method that applies focus measures on original image planes.     

结合注意力的3D卷积网络脑胶质瘤分割算法

为了提升脑胶质瘤分割精度，提出一种结合注意力机制的3D卷积神经网络算法。输入3个不同尺度的图像块，经过9个卷积层和1个分类层后得到3个不同的分类结果，将分类结果与注意力学习到的权重相乘并逐体素相加得到输出。此外该算法采用了一种混合Dice损失函数与Focal损失函数的超参数损失函数。实验表明，该算法的Dice系数在整体区域、核心区域以及增强区域分别达到了95.31%、80.12%、82.25%。与已有的一种脑胶质瘤分割算法deepmedic相比，整体区域、核心区域以及增强区域的Dice系数分别提升了3%、2%、6%。在脑胶质瘤分割方面，具有重要的临床意义。     

基于卷积神经网络的单目深度估计

针对利用深度学习方法对街道图像进行深度估计，提出采用语义分割的方法解决深度图出现边界模糊等问题；估计深度通过左右视角图生成视差图进行无监督的训练。在网络模型中添加语义分割层，采取多个空洞卷积并行的结构增加感受野，同时减少了图像下采样的次数，降低了由于下采样带来的信息损失，使得的结果更加准确。这也是在深度估计中首次与空洞卷积相结合增加准确率。通过对KITTI街道数据集进行训练，与现有结果相比，除了增加检测准确性，降低错误率之外，使得效果图中的物体更加清晰，并且在效果图中还保留了一些原模型中被忽视掉的细节信息，将原始图像更加完整的表现出来。     

Benchmark value determination of energy efficiency indexes for coal-fired power units based on data mining methods

The operational optimisation of coal-fired power units is important for saving energy and reducing losses in the electric power industry. One of the key issues is how to determine the benchmark values of the energy efficiency indexes of the units. Therefore, a new framework for determining these benchmark values is proposed, based on data mining methods. First, the energy efficiency key performance indicators (KPIs) associated with the net coal consumption rate (NCCR) were selected based on the domain knowledge. Second, the decision-making samples with minimal NCCR were acquired with the fuzzy C-means (FCM) clustering algorithm, and the corresponding clustering centres were employed as the benchmark values. Finally, based on the support vector regression (SVR) algorithm, the target values of the NCCR were obtained with the KPIs as input, and the energy saving potential was evaluated by comparing the target values with the historical values of the NCCR. An actual on-duty 1000 MW unit was taken as study unit, and the results show that the energy saving potential is remarkable when the operators adjust the KPIs based on the calculated benchmark values.     

基于改进奇异值分解滤波和谱峭度的滚动轴承故障诊断

针对含噪信号的有效奇异值个数难以确定的问题，提出了一种改进的奇异值分解降噪方法--奇异值累积法。该方法通过计算奇异值的实际下降值与奇异值平均下降速度累积量的差值，并取该差值最大值点的位置作为有效奇异值的分界点来确定有效奇异值的个数。在此基础上，提出了一种基于奇异值累积法与快速谱峭度的滚动轴承故障诊断方法。采用奇异值累积法对原信号进行降噪处理，然后利用快速谱峭度确定滤波器中心频率及带宽，通过分析频段包络谱中明显的频率成分来诊断故障。该方法可以有效去除信号中的噪声，使得到的峭度值所反映的故障冲击更接近实际情况。对含内圈、外圈故障的滚动轴承实验数据进行分析，实验结果表明，相比快速谱峭度的故障诊断方法，该方法具有更好的故障识别效果。     

Co-creation of value and the project context: Towards application on the case of Hinkley Point C Nuclear Power Station

The research analyses decisions as evaluative outcomes regarding project value. The UK-French Hinkley Point C Nuclear Power Station provides the case study. Value is traditionally assessed as inputs and outputs. Here, value is conceptualized as a co-created value proposition at the front-end with implications for realization post-completion. Service-dominant logic (SDL) provides the theoretical lens and contributes to a methodological approach for examining projects. Decision-making outcomes provide the evidence from a range of data sources, including reports and commentaries. The methods are interpretative. The findings show that decision-making extends beyond the time-cost-quality/scope dimensions. The long-term issues regarding value realized are often overlooked. Stakeholders and individual actors have mainly focused upon managing political and financial risks, especially time and cost. The research poses challenges to project management analysis, SDL and research design in assessing evidence. Addressing these issues facilitates a knowledge contribution to SDL theorization and the field of project management.     

Ex post demand information sharing between differentiated suppliers and a common retailer

We consider ex post demand information sharing in a two-echelon supply chain in which two suppliers sell differentiated products through a common retailer. We model three scenarios of information sharing to characterise the conditions under which information sharing may benefit or hurt different players: (1) no supplier is informed; (2) only one supplier is informed; and (3) both suppliers are informed. We find that the retailer may voluntarily share the low demand information when the product differentiation and demand magnitude satisfy certain conditions. In contrast, the retailer has no incentive to share the high demand information, even though this information benefits the suppliers. To achieve information sharing, we propose a side payment mechanism, which allows benefitted players to subsidise hurtful players. Finally, we extend our model by considering a situation where the suppliers have limited capacity and may invest to ramp up production capacity to satisfy a high demand.     

An expectation operator for belief functions in the Dempster–Shafer theory*

ABSTRACT

The main contribution of this paper is a new definition of expected value of belief functions in the Dempster–Shafer (D–S) theory of evidence. Our definition shares many of the properties of the expectation operator in probability theory. Also, for Bayesian belief functions, our definition provides the same expected value as the probabilistic expectation operator. A traditional method of computing expected of real-valued functions is to first transform a D–S belief function to a corresponding probability mass function, and then use the expectation operator for probability mass functions. Transforming a belief function to a probability function involves loss of information. Our expectation operator works directly with D–S belief functions. Another definition is using Choquet integration, which assumes belief functions are credal sets, i.e. convex sets of probability mass functions. Credal sets semantics are incompatible with Dempster's combination rule, the center-piece of the D–S theory. In general, our definition provides different expected values than, e.g. if we use probabilistic expectation using the pignistic transform or the plausibility transform of a belief function. Using our definition of expectation, we provide new definitions of variance, covariance, correlation, and other higher moments and describe their properties.