基于改进Wasserstein生成式对抗网络的电力系统不良数据辨识

随着新能源并网以及大量电力电子器件的投入,电力系统的数据类型向多元复杂化的趋势发展。针对大规模电力系统中出现的不良数据辨识性能差、辨识效率低的问题,提出了一种基于改进Wasserstein生成式对抗网络(WGAN-GP)的不良数据辨识方法。基于历史数据库中的状态量得到多断面正常量测数据并训练WGAN-GP模型;将含不良数据的量测信息输入训练好的WGAN-GP模型,得到对应的量测重构数据,并得到最终的量测重构误差;为了避免人为确定阈值的主观性,提出了一种基于C4.5决策树模型的不良数据阈值确定方法,将量测重构误差输入训练好的决策树模型,即可定位1组量测信息中的不良数据位置。以IEEE标准系统和某实际省网为算例进行仿真测试,结果表明相较于已有方法,所提方法具有更好的辨识性能和更高的辨识效率。     

平行视觉:基于ACP的智能视觉计算方法

在视觉计算研究中,对复杂环境的适应能力通常决定了算法能否实际应用,已经成为该领域的研究焦点之一.由人工社会（Artificial societies）、计算实验（Computational experiments）、平行执行（Parallel execution）构成的ACP理论在复杂系统建模与调控中发挥着重要作用.本文将ACP理论引入智能视觉计算领域,提出平行视觉的基本框架与关键技术.平行视觉利用人工场景来模拟和表示复杂挑战的实际场景,通过计算实验进行各种视觉模型的训练与评估,最后借助平行执行来在线优化视觉系统,实现对复杂环境的智能感知与理解.这一虚实互动的视觉计算方法结合了计算机图形学、虚拟现实、机器学习、知识自动化等技术,是视觉系统走向应用的有效途径和自然选择.     

一类基于类别区分的多重故障诊断模型框架

多重故障诊断是故障诊断和容错控制技术中的重难点问题.本文关注数据驱动故障诊断,以基于知识的方法为基本思路,按照故障表现及征兆的组合形式分析了多重故障和征兆之间的映射关系,将多重故障诊断转化为一类由各个组成故障对应于征兆数据集合的类划分问题.在已有应用于故障诊断的分类和聚类方法的基础上,分析了4种基于类别区分的多重故障诊断模型框架.并讨论了其优缺点和适用的类别区分算法.     

数据驱动的锑粗选泡沫图像特征优化设定

针对锑浮选过程中精、尾矿品位难以在线检测, 浮选性能不稳定的问题, 提出一种数据驱动的泡沫图像特征优化设定方法. 该方法根据入矿品位类型对泡沫图像特征进行优化设定, 并针对不同入矿品位类型的样本分布特点,先尝试采用案例推理的方法从历史数据中寻找浮选性能优良的泡沫状态. 若经验知识不足, 则采用基于多中心模糊C均值聚类与概率支持向量回归的区间II 型模糊系统建模方法建立精、尾矿品位指标模型, 并在此基础上利用智能优化方法寻优泡沫图像特征值. 某锑浮选工业实验结果表明了所提出方法的有效性.

     

Scalable Linear Array Architecture with Data-Driven Control for Ultrahigh-Speed Vector Quantization

Current and future requirements for adaptive real-time image compression challenge even the capabilities of highly parallel realizations in terms of hardware performance. Previously proposed linear array structures for full-search vector quantization do not offer scalability and adaptivity in this context, because they require separate data/control pins for dynamically updating the codevectors and complicated interlock mechanisms to ensure that the regular data flow is not corrupted as a result of updates. We explore the design space for full-search vector quantizers and propose a novel linear processor array architecture in which global wiring is limited to clock and power supply distribution, thus allowing high-speed processing in spite of only limited communication with the host via the boundary processors. The resulting fully pipelined design is not only area-efficient for VLSI implementation but is also readily scalable and offers extremely high performance.     

Improved stochastic subspace method for identifying structural modal parameters

Stochastic subspace identification can be used to identify the modal parameters of a structure according to its dynamic response to ambient excitation. However, some high dimensional matrices (Toeplitz matrices) must be constructed in the process of identification, and lots of memory and computation time are cost to the singular value decomposition of these high dimensional matrixes. Stochastic subspace identification affects the computational efficiency seriously. Therefore, this paper investigates a new method for constructing lower-dimension Toeplitz matrices to improve the computing efficiency. Finally, a numerical simulation is presented to demonstrate the computing efficiency of the method. The result shows that the computing consumption of the proposed method is only 106％ the computing consumption of the traditional stochastic subspace identification while the identification accuracy is maintained.     

Proper Generalized Decomposition based dynamic data driven inverse identification

Dynamic Data-Driven Application Systems—DDDAS—appear as a new paradigm in the field of applied sciences and engineering, and in particular in Simulation-based Engineering Sciences. By DDDAS we mean a set of techniques that allow the linkage of simulation tools with measurement devices for real-time control of systems and processes. One essential feature of DDDAS is the ability to dynamically incorporate additional data into an executing application, and in reverse, the ability of an application to dynamically control the measurement process. DDDAS need accurate and fast simulation tools using if possible off-line computations to limit as much as possible the on-line computations. With this aim, efficient solvers can be constructed by introducing all the sources of variability as extra-coordinates in order to solve the model off-line only once. This way, its most general solution is obtained and therefore it can be then considered in on-line purposes. So to speak, we introduce a physics-based meta-modeling technique without the need for prior computer experiments. However, such models, that must be solved off-line, are defined in highly multidimensional spaces suffering the so-called curse of dimensionality. We proposed recently a technique, the Proper Generalized Decomposition—PGD—able to circumvent the redoubtable curse of dimensionality. The marriage of DDDAS concepts and tools and PGD off-line computations could open unimaginable possibilities in the field of dynamic data-driven application systems. In this work we explore some possibilities in the context of on-line parameter estimation.     

Materials inspired by mathematics

Our world is transforming into an interacting system of the physical world and the digital world. What will be the materials science in the new era? With the rising expectations of the rapid development of computers, information science and mathematical science including statistics and probability theory, ‘data-driven materials design’ has become a common term. There is knowledge and experience gained in the physical world in the form of know-how and recipes for the creation of material. An important key is how we establish vocabulary and grammar to translate them into the language of the digital world. In this article, we outline how materials science develops when it encounters mathematics, showing some emerging directions.