Sequential decision making in general state space models

Many problems in control and signal processing can be formulated as sequential decision problems for general state space models. However, except for some simple models one cannot obtain analytical solutions and has to resort to approximation. In this thesis, we have investigated problems where Sequential Monte Carlo (SMC) methods can be combined with a gradient based search to provide solutions to online optimisation problems. We summarise the main contributions of the thesis as follows. Chapter 4 focuses on solving the sensor scheduling problem when cast as a controlled Hidden Markov Model. We consider the case in which the state, observation and action spaces are continuous. This general case is important as it is the natural framework for many applications. In sensor scheduling, our aim is to minimise the variance of the estimation error of the hidden state with respect to the action sequence. We present a novel SMC method that uses a stochastic gradient algorithm to find optimal actions. This is in contrast to existing works in the literature that only solve approximations to the original problem. In Chapter 5 we presented how an SMC can be used to solve a risk sensitive control problem. We adopt the use of the Feynman-Kac representation of a controlled Markov chain flow and exploit the properties of the logarithmic Lyapunov exponent, which lead to a policy gradient solution for the parameterised problem. The resulting SMC algorithm follows a similar structure with the Recursive Maximum Likelihood(RML) algorithm for online parameter estimation. In Chapters 6, 7 and 8, dynamic Graphical models were combined with with state space models for the purpose of online decentralised inference. We have concentrated more on the distributed parameter estimation problem using two Maximum Likelihood techniques, namely Recursive Maximum Likelihood (RML) and Expectation Maximization (EM). The resulting algorithms can be interpreted as an extension of the Belief Propagation (BP) algorithm to compute likelihood gradients. In order to design an SMC algorithm, in Chapter 8 uses a nonparametric approximations for Belief Propagation. The algorithms were successfully applied to solve the sensor localisation problem for sensor networks of small and medium size.

Air flow over a mountain and the convective boundary-layer

In this paper, the analytical model coupling the convective boundary layer (CBL) with the free atmosphere developed by Qi and Fu (1992) is improved. And by this improved model, the interaction between airflow over a mountain and the CBL is further discussed. The conclusions demonstrate: (1) The perturbation potential temperatures in the free atmosphere can counteract the effect of orographic thermal forcing through entraining and mixing in the CBL. If u(M)BAR > u(F)BAR, the feedback of the perturbation potential temperatures in the free atmosphere is more important than orographic thermal forcing, which promotes the effect of interfacial waves. If u(M)BAR < u(F)BAR, orographic thermal forcing is more important, which makes the interfacial height and the topographic height identical in phase, and the horizontal speeds are a maximum at the top of the mountain. (2) The internal gravity waves propagating vertically in the free atmosphere cause a strong downslope wind to become established above the lee slope in the CBL and result in the hydraulic jump at the top of the CBL. (3) With the CBL deepening, the interfacial gravity waves induced by the potential temperature jump at the top of the CBL cause the airflow in the CBL to be subcritical.

Magnetic properties of sintered sm2fe17ny magnets

Sintered magnets of Sm2Fe17Ny nitrides, with a density of 6.0-7.4 g/cm3, have been prepared by using an explosion technique. Both crystalline structure and the magnetic properties of Sm2Fe17Ny nitrides were retained in the process. The sintered magnet had a remanence B(r)=0.83 T, an intrinsic coercivity mu(0i)H(c)=0.57 T and an energy product (BH)max=88 kJ/m3. The temperature dependence of coercivity and remanence were also measured. The temperature coefficients alpha of remanence and beta of coercivity are -0.076%/degrees-C and -0.51%/degrees-C, respectively.

基于人工神经网络的油水旋流分离器分离性能预测

采用改进的BP神经网络模型模拟水力旋流器的油水分离过程.根据水力旋流器的实际运行条件,确定旋流器模型设计中的优化神经网络结构,将遗传算法用于优化三层BP神经网络的初始权重,采用PRP共轭梯度法优化BP算法.结果表明,采用人工神经网络模型预测油水分离水力旋流器的分离性能是切实可行的,它能成功地模拟旋流器的分离过程,进而实现旋流器操作控制的优化.

利用改进的神经网络预测腐蚀管道的剩余强度

将BP神经网络和遗传算法相结合 ,得到一种新的神经网络 ,并将这种神经网络成功用于计算腐蚀管道的剩余强度和最大允许输送压力。通过示例分析 ,得到下面结论 :不同计算方法计算得到的剩余强度和最大允许输送压力相差较大 ,Wes - 2 80 5 - 97规范、ASME -B31G规范、CVDA— 84规范等都比J积分方法计算得到的剩余强度和最大允许输送压力偏大 ;DM断裂力学方法计算得到的剩余强度和最大允许输送压力比J积分偏小 ;J积分方法和基于J积分方法的改进的遗传神经网络方法计算结果比较接近 ,可以认为是计算剩余强度和最大允许输送压力较好的方法

人工神经网络技术在成形模具表面激光强化中的应用

基于人工神经网络基本理论，建立在成形模具表面激光强化中预测材料表面最高温度、强化效果的BP网络模型。借助于MATLAB仿真软件中的神经网络工具箱作为开发平台，将试验样本数据和经过试验验证的数值计算结果作为补充的样本数据用于BP网络的训练，利用训练好的BP网络对非线性的样本数据规律进行拟合，实现激光加工工艺参数的优化，为实际生产和加工提供有效的依据。

Financing And Cooperation For Sustainable Development In The Mediterranean, Sophia Antipolis, France, 3-4 June 2004: Volume II: Technical Report of the Workshop

(263 page document)

预测腐蚀管道剩余强度的新方法

将BP神经网络和遗传算法相结合 ,得到了一种可计算腐蚀管道剩余强度和最大允许注水压力的新神经网络。通过实例分析 ,将 7种常用规范和改进的遗传神经网络方法进行了比较。结果表明 ,不同计算方法得到的剩余强度和最大允许注水压力相差较大 ,Wes 2 80 5 - 97规范、ASMEB31G规范、CVDA— 84规范、Irwin断裂力学方法等都比J积分方法的剩余强度和最大允许注水压力偏大 ;Burdiken和DM断裂力学方法计算得到的剩余强度和最大允许注水压力比J积分偏小 ;J积分方法和基于J积分方法的改进遗传神经网络方法计算结果比较接近 ,比较适中 ,可以认为是计算管道剩余强度和最大允许注水压力较好的方法

Shell midden people in northern Iberia. New data from the mesolithic rock shelter of J3 (Basque Country, Spain)

[EN]In the course of a sondage dug in the rock shelter of J3, in the Jaizkibel mountains (at the north-western tip of Guipúzcoa), the body of a adult man was located buried inside a shell midden. This shell midden had not been disturbed and presented internal stratigraphy features. In any case, the outer edge of the shell midden does show some interesting interdigitation with the adjacent habitational layers, with evidence of different stages of occupation. Within the shell midden itself, under the individual buried there, it was possible to observe layers without any ceramics, whereas the layers covering said individual included ceramic fragments. This individual has been dated to 8,300 BP and therefore corresponds to a Mesolithic context.

腐蚀管道的剩余强度计算方法研究

对目前国内外常用规范进行归纳和总结,得到5种常用方法: ASME—B31G,DM方法,Wes-2805—97,CVDA-84和J积分方法.本文将BP神经网络和遗传算法相结合,得到一种新的神经网络,并将这种神经网络成功用于计算腐蚀管道的剩余强度和最大允许注水压力.通过示例分析,对7种常用规范和本文提出的改进的遗传神经网络方法进行了比较,得到下面结论:不同计算方法计算得到的剩余强度和最大允许注水压力相差较大, Wes-2805-97规范、 ASME—B31G规范、 CVDA-84规范等都比J积分方法计算得到的剩余强度和最大允许注水压力偏大、偏保守; DM断裂力学方法计算得到的剩余强度和最大允许注水压力比J积分偏小、偏危险; J积分方法和基于J积分方法的改进的遗传神经网络方法计算结果比较接近,比较适中,可以认为是计算剩余强度和最大允许注水压力较好的方法.

Adaptación de una aplicación RIA desarrollada en Flex a una aplicación HTML5

Babelium Project (tambi en BP de aqu en adelante), es un proyecto de c odigo abierto cuyo principal objetivo es fomentar el aprendizaje de idiomas online. Para ello, haciendo uso de distintas tecnolog as, se cre o una aplicaci on web, Rich Internet Applications (RIA), que permite a los usuarios practicar idiomas por pr actica oral. La aplicaci on dispone de varios m odulos donde, gracias al streaming de v deo, los usuarios pueden grabar o evaluar ejercicios de forma colaborativa. Babelium Project empez o siendo una aplicaci on desarrollada bajo el conjunto de tecnolog as Flex, un conjunto de tecnolog as de Adobe cuya compilaci on resultante es un aplicaci on (web en este caso) basada en Flash. Sin embargo, con la reciente llegada de HTML5, todo parece indicar que Adobe abandonar a Flex en un futuro 1 2 para centrar sus esfuerzos en el desarrollo de soluciones para esta nueva tecnolog a. Por esa raz on, naci o este proyecto, con el n de migrar Babelium Project a HTML5, conjunto de tecnolog as con gran futuro y acogida en el mundo web. Mi trabajo en este proyecto ha consistido, principalmente, en analizar la factibilidad y proceso adecuado de la migraci on de un proyecto de dimensiones considerables desarrollado bajo Flex al conjunto de tecnolog as que forman HTML5, teniendo para ello el proyecto Babelium como caso de prueba. Las fases principales del proyecto han sido: an alisis del estado de HTML5, an alisis de factibilidad, elecci on de un conjunto de tecnolog as para la migraci on, desarrollo de patrones de migraci on y, por ultimo, migraci on de Babelium utilizando dichas tecnolog as y siguiendo dichos patrones.

Association analyses of DNA polymorphisms in bovine SREBP-1, LXRα,FADS1 genes with fatty acid composition in Canadian commercial crossbred beef steers

Two previously reported DNA polymorphisms of sterol regulatory element binding transcription factor 1 (SREBP1) and liver X receptor alpha (LXRα) and two DNA polymorphisms of fatty acid desaturase 1 (FADS1) were evaluated for associations with fatty acids in brisket adipose tissue of Canadian cross-bred beef steers. The polymorphism of 84 bp insert/deletion in intron 5 of SREBP1 was significantly associated with the concentration of 9c C17:1 (P=0.013). The G>A single nucleotide polymorphism (SNP) in the exon 4 of LXRα gene was associated with the concentration of 9c, 11t C18:2 (P=0.04), sum of conjugated linoleic acids (CLA) (P=0.025) and 11c C20:1(P=0.042). Two DNA polymorphisms in the promoter region of FADS1, deletion/insertion of ->GTG in rs133053720 and SNP A>G in rs42187276, were significantly associated with concentrations of C17:0 iso, C17:0 ai, total branched chain fatty acids (BFA), 12t C18:1, 13t/14t C18:1, 15t C18:1, and 13c C18:1 (P<0.05). Further studies are needed to validate the associations and to delineate the roles of the gene polymorphisms in determining the fatty acid composition in beef tissues.

Analytical Methods of Residual Strength in Injecting Water Pipeline

In this paper, common criterions about residual strength evaluation at home and abroad are generalized and seven methods are acquired, namely ASME-B31G, DM, Wes-2805-97, CVDA-84, Burdekin, Irwin and J integral methods. BP neural network are Combined with Genetic Algorithm (GA) named by modified BP-GA methods to successfully predict residual strength and critical pressure of injecting water, corrosion pipelines. Examples are shown that calculation results of every kind of method have great difference and calculating values of Wes-2805-97 criterion, ASME-B31G criterion, CVDA-84 criterion and Irwin fracture mechanics model are conservative and higher than, those of J integral methods while calculating values of Burdiken model and DM fracture mechanics model are dangerous and less than those of J integral methods and calculating values of modified BP-GA methods are close and moderate to those of J integral methods. Therefore modified BP-GA methods and J integral methods are considered better methods to calculate residual strength and critical pressure of injecting water corrosion pipelines