基于模糊推理的多智能体强化学习

以电子市场智能定价问题为研究背景,提出基于模糊推理的多智能体强化学习算法(FI-MARL).在马尔科夫博弈学习框架下,将领域知识初始化为一个模糊规则集合,智能体基于模糊规则选择动作,并采用强化学习来强化模糊规则.该方法有效融合应用背景的领域知识,充分利用样本信息并降低学习空间维数,从而增强在线学习性能.在电子市场定价的...     

Phase change materials in the ternary system NH4Cl+CaCl2+H2O

Solubility isotherms of the ternary system (NH₄Cl+CaCl₂+H₂O) were elaborately determined at T= (273.15 and 298.15) K by using the isothermal method. In the equilibrium phase diagram, there are two solubility branches corresponding to the solid phases CaCl₂⋅6H₂O and NH₄Cl. Invariant point compositions are 36.32 wt% CaCl₂ and 3.4 wt% NH₄Cl at 273.15 K, and 45.86 wt% CaCl₂ and 5.22 wt% NH₄Cl at 298.15 K. A Pitzer-Simonson-Clegg thermodynamic model was applied to represent the thermodynamic properties of this ternary system and to construct a partial phase diagram of the ternary system at temperatures between (273.15 and 323.15) K. It was found in the predicted solubility phase diagram that the double salt 2NH₄Cl⋅CaCl₂⋅3H₂O, found by other authors at (323.1 and 348.1) K, will disappear at temperatures below 298.15 K. Besides, it was found that there are two peritectic points in the ternary system with peritectic temperatures at 299.65 K and 298.15 K, and the former peritectic point falls just on the line between the composition points of NH₄Cl and CaCl₂⋅6H₂O. According to phase rule, a solution made of this point will begin to crystallize at 299.65 K and end at 298 K and therefore can be acted as a “pseudo eutectic” phase change material (PCM). A heat storing and releasing experiment of 50 grams of the PCM was carried out, obtaining a satisfying result.     

Product structure (BOM)-based product similarity measures using orthogonal procrustes approach

In geometry, two configurations are called similar if they can be brought to a complete match by rigid motions and dilations. In the simplest case, product designs X and Y have the same dimensionality and same number of points which can be brought into a one to one correspondence by substantive considerations. Under orthogonal transformations, product design Y can only be rotated and reflected in order to approximate product design X. As for the “as close as possible” or “relatively close” criterion, a reasonable definition would be to measure the distances between corresponding points, square these values, and add them to obtain the sum-of-squares criterion L. In this paper, a product structure (BOM)-based product similarity measures using orthogonal procrustes approach is proposed. A structural product is first transformed into bill of materials (BOM), the structural similarity measurement between a product query and target product is calculated and evaluated using orthogonal procrustes approach. The results show that the proposed product structure similarity measure method can help companies find desired product information based on BOM feature.     

Incorporating haptic feedback in simulation for learning physics

The purpose of this study was to investigate the effectiveness of a haptic augmented simulation in learning physics. The results indicate that haptic augmented simulations, both the force and kinesthetic and the purely kinesthetic simulations, were more effective than the equivalent non-haptic simulation in providing perceptual experiences and helping elementary students create multimodal representations of the movements of gears. However, in most cases, force feedback was needed to construct a fully loaded multimodal representation that helps students to comprehend later instruction with less sensory modalities. In addition, the force and kinesthetic simulation was effective in helping to transfer knowledge to new learning situations. These findings suggest that it is important to help elementary students make a solid cognitive grounding with the use of a perceptual anchor.     

Region-based parallelization of irregular reductions on explicitly managed memory hierarchies

Multicore architectures are evolving with the promise of extreme performance for the classes of applications that require high performance and large bandwidth of memory. Irregular reduction is one of important computation patterns for many complex scientific applications, and it typically requires high performance and large bandwidth of memory. In this article, we propose region-based parallelization techniques for irregular reductions on multicore architectures with explicitly managed memory hierarchies. Managing memory hierarchy in software requires a lot of programming efforts and tends to be error-prone. The difficulties are even worse for applications with irregular data access patterns. To relieve the burden of memory management from programmers, we develop abstractions, particularly targeted to irregular reduction, for structuring parallel tasks, mapping the parallel tasks to processing units and scheduling data transfers between the memory hierarchies. Our framework employs iteration reordering based on regions of data along with dynamic scheduling of parallel tasks. We experimentally evaluate the effectiveness of our techniques for irregular reduction kernels on the Cell processor embedded in a Sony PlayStation3. Experimental results show the speedups of 8 to 14 on the six available SPEs.     

Geolocation of Multiple Targets from Airborne Video Without Terrain Data

The task of geolocating targets from airborne video is required for many applications in surveillance, law enforcement, reconnaissance, etc. The usual approaches to target geolocation involve terrain data, single target tracking, gimbal control of camera heads, altimeters, etc. The main goal of this research is to eliminate those requirements and still develop an accurate, efficient, and robust vision-based method for geolocation that can be carried out for multiple targets simultaneously. In that sense, our main contributions to the state-of-the-art in geolocation are fourfold: 1) to eliminate the requirement for gimbal control of the cameras or any particular path planning control for the UAV; 2) to perform instaneous geolocation of multiple targets; 3) to eliminate the requirements for geo-referenced terrain database (elevation maps) or for an altimeter that provides the UAV’s and target’s altitudes; and 4) to use one single camera while still maintaining good overall accuracy. In order to achieve that, the only requirements for our proposed method are: that the intrinsic parameters of the camera be known; that the on board camera be equipped with global positioning system (GPS) and inertial measurement unit (IMU); and that the height of the vehicle can be calculated using feature points extracted from the ground surrounding the image of the targets. To satisfy the first two requirements, we developed and tested a robust calibration procedure that can estimate not only the intrinsic parameters of the camera, but also the IMU-camera parameters (also know in the robotic circles as the hand-eye calibration). The last requirement was addressed using a pseudo-stereo vision technique that maximizes the distance between stereo pairs (baseline) while keeping large the number of common feature points extracted by the algorithm. The result is a method that can reach approximately 25 m of accuracy for an UAV flying at 155 m away from the target. Such performance is demonstrated by computer simulation, in-scale data using a model city, and real airborne video with ground truth.     

Multiwalled carbon nanotubes grafted chitosan nanobiocomposite: A prosperous functional nanomaterials for glucose biosensor application

Multiwalled carbon nanotubes (MWNTs) grafted chitosan (CS) nanowire (NW) was prepared by phase separation method. Glucose oxidase (GOx) was sequentially immobilized into MWNT-CS-NW to obtain MWNT-CS-NW/GOx biosensor. Field emission scanning electron microscopy (FESEM) images of MWNT-CS-NW/GOx reveals the existence of MWNT and CS. Cyclic voltammetry and amperometry were used to evaluate the electrochemical determination of glucose. The MWNT-CS-NW/GOx biosensor shows an excellent performance for glucose at +0.34 V with a high sensitivity (5.03 μA/mM) and lower response time (3 s) in a wide concentration range of 1-10 mM (correlation coefficient of 0.9988). In addition, MWNT-CS-NW/GOx biosensor possesses better reproducibility, storage stability and there is negligible interference from other electroactive components.     

Nano-structured solid oxide fuel cell design with superior power output at high and intermediate operation temperatures

A solid oxide fuel cell (SOFC) with a thin-film yttria-stabilized zirconia (YSZ) electrolyte was developed and tested. This novel SOFC shows a similar multilayer set-up as other current anode-supported SOFCs and is composed of a Ni/8YSZ anode, a gas-tight 8YSZ electrolyte layer, a dense Sr-diffusion barrier layer and a LSCF cathode. To increase the power density and lower the SOFC operating temperature, the thickness of the electrolyte layer was reduced from around 10 μm in current cells to 1 μm, using a nanoparticle deposition method. By using the novel 1 μm electrolyte layer, the current density of our SOFC progressed to 2.7, 2.1 and 1.6 A/cm² at operation temperatures of 800, 700 and 650°C, respectively, and out-performs all similar cells reported to date in the literature. An important consideration is also that cost-effective dip-coating and spin-coating methods are applied for the fabrication of the thin-film electrolyte. Processing of 1 μm layers on the very porous anode substrate material was initially experienced as very difficult and therefore 8YSZ nanoparticle coatings were developed and optimized on porous 8YSZ model substrates and transferred afterwards to regular anode substrates. In this paper, the preparation of the novel SOFC is shown and its morphology is illustrated with high resolution SEM pictures. Further, the performance in a standard SOFC test is demonstrated.