Search

Wrapper-based feature selection approaches reliant on different bio-inspired optimisation algorithms are both effective and widely employed when dealing with classification problems. These algorithms have proven themselves as successful wrappers in finding good feature subsets. However, as a large number of them is defined for the real domain, the small detail of their adaptation to the discrete domain of feature selection is often overlooked. This holds especially true for differential evolution, a prominent wrapper choice among bioinspired optimisation algorithms. As distinct discretisation techniques have been proposed in the literature, the question of which one to incorporate in differential evolution and under which circumstances remains rather unanswered. This paper attempts to provide some answers in that regard by studying the incorporation of discretisation techniques into differential evolution and their influence on the quality of attained feature subsets. Given their differences, some suggestions concerning the selection of discretisation techniques are given based on the obtained results.

23rd International Conference on Knowledge-Based and Intelligent Information & Engineering Systems - early article, The paper presents research on unsupervised and supervised discretisation of input data used in execution of stylometric tasks of authorship attribution. Basing on numeric characterisation of writing styles, recognition of authorship is performed by decision rules, as their transparent structure enhances understanding of discovered knowledge. The performance of rule classifiers, constructed in rough set approach, is studied in the context of a strategy employed for resolving conflicts. It is also contrasted with that of other selected inducers.

peer-reviewed Improving building energy efficiency is of paramount importance due to the large proportion of energy consumed by thermal operations. Consequently, simulating a building's environment has gained popularity for assessing thermal comfort and design. The extended timeframes and large physical scales involved necessitate compact modelling approaches. The accuracy of such simulations is of chief concern, yet there is little guidance offered on achieving accurate solutions whilst mitigating prohibitive computational costs. Therefore, the present study addresses this deficit by providing clear guidance on discretisation levels required for achieving accurate but computationally inexpensive models. This is achieved by comparing numerical models of varying discretisation levels to benchmark analytical solutions with prediction accuracy assessed and reported in terms of governing dimensionless parameters, Biot and Fourier numbers, to ensure generality of findings. Furthermore, spatial and temporal discretisation errors are separated and assessed independently. Contour plots are presented to intuitively determine the optimal discretisation levels and time-steps required to achieve accurate thermal response predictions. Simulations derived from these contour plots were tested against various building conditions with excellent agreement observed throughout. Additionally, various scenarios are highlighted where the classical single lumped capacitance model can be applied for Biot numbers much greater than 0.1 without reducing accuracy.

A Rotationally Symmetric Spherical Discretisation (RSSD) technique has recently been developed for the discretisation of scattering phase functions in the Discrete Ordinates Method (DOM) in a consistent and rational manner. RSSD has inherent energy conservation, is suitable for any quadrature scheme used, minimising run-time matrix calculation, and to date has been used for the S4, S6 and S8 ordinate sets. In this follow-up paper, the RSSD boundary intervals for the four weightings of the S10 ordinate set are presented. The effect of asymmetry factor g and angular resolution of the discrete input scattering phase function distribution “grid” on final RSSD errors (in scattered energy conservation and calculated asymmetry factor) is explored, making use of the Henyey–Greenstein (HG) family of distributions. It is shown that the RSSD scattered energy error View the MathML sourceξeRSSD is a strong function of the grid resolution, indicating that View the MathML sourceξeRSSD can be controlled down to user-required levels by judicious choice of grid resolution and by local grid refinement in regions of steep gradients and peaks in the scattering phase function distribution (in this paper error levels as low as 0.002% have been achieved). RSSD asymmetry factor error View the MathML sourceξgRSSD behaves differently to View the MathML sourceξeRSSD, however, displaying a strong grid resolution dependence in a “grid sensitivity zone” of combinations of coarse grid resolution and high values of prescribed g (for the HG family of scattering phase functions, above g values of 0.9, 0.95 and 0.98 for grid angular resolutions of 1°, 0.5° and 0.25° respectively). Outside of the “grid sensitivity zone” g error is insensitive to grid resolution and achieves maximum (positive and negative) values of between −2.7% and 1% between prescribed g values of 0.7 and 0.9. This band decreases to a range from −1% to 1% for the S4, S6 and S10 ordinate sets. The grid resolution conclusions are supported by examining RSSD discretisation for two real scattering phase function distributions: for a diffusely reflecting large sphere with an asymmetry factor of −0.4444, grid angular resolutions of 2°, 1°, 0.5° and 0.25° resulted in the decreasing maximum View the MathML sourceξeRSSD values of 0.0449%, 0.0136%, −0.0050% and −0.0015% respectively, but near-constant maximum View the MathML sourceξgRSSD values of −1.225%, −1.239%, −1.247% and −1.250% respectively, while for a transparent large sphere with an asymmetry factor of 0.660, grid angular resolutions of 1°, 0.5° and composite 0.5°/0.25° resolution resulted in the decreasing maximum View the MathML sourceξeRSSD values of 0.0685%, −0.0402% and −0.0046% respectively and near-constant maximum View the MathML sourceξgRSSD values of −1.286%, −1.310% and −1.313% respectively

Multiple uncertainties complicate socio-economic forecasting problems, especially when relying on ill-conditioned limited data. Such problems are best addressed by grey prediction models such as Grey Verhulst Model (GVM). This paper resolves the incompatibility between GVM’s estimation and prediction by taking its basic form equation as the basis of both. The resultant “Basic Form”-focused GVM (BFGVM) is also further developed to create Direct Non-equidistant BFGVM (DNBFGVM) and, in turn, DNBFGVM with Recursive simulation (DNBFGVMR). Experimental analyses comprise 19 socio-economic time series with an emphasis on Iranian population, a low-frequency non-equidistant time series with remarkable strategic importance. Promisingly, the proposed DNBFGVM and DNBFGVMR provide accurate in-sample and out-of-sample socio-economic forecasts, show highly significant improvements over the best traditional GVM, and offer cost-effective intelligent support of decision-making. Final results suggest future trends of studied socio-economic time series. Specifically, they reveal Iranian population to grow even slower than anticipated, demanding an urgent consideration of policy-makers.

Heat transfer in wall-flow monoliths has gained in interest because of the widespread adoption of these systems by automotive industry to fulfil soot emission regulations and the importance of heat exchange on the regeneration process control to avoid damaging the monolith. This paper presents a heat transfer model for wall-flow diesel particulate filters coupled with an unsteady compressible flow solver. The heat exchange between the gas and the solid phase is based on a bi-dimensional discretisation of the porous medium both in axial and tangential directions. The monolith can be discretised in the radial direction to account for the heat fluxes towards the environment through the monolith and the canister, which is also coupled with the inlet and outlet ducts of the filter. The model is validated against experimental data obtained in a flow test rig. A test campaign under non-reacting conditions has been conducted to show the capability for thermal response prediction. Tests cover clean and soot loaded monolith, continuous flow under steady and transient thermal conditions, and pulsating flow. In this case, the characteristics of the pressure waves in amplitude and frequency are similar to those that the monolith can undergo depending on its location along the exhaust line. © 2012 Elsevier Ltd., This work has been partially supported by the Spanish Ministerio de Ciencia e Innovacion through grant number DPI2010-20891-C02-02.

This research explores the photofermentation of glycerol to hydrogen by Rhodopseudomonas palustris, with the objective to maximise hydrogen production. Two piecewise models are designed to simulate the entire growth phase of R. palustris; a challenge that few dynamic models can accomplish. The parameters in both models were fitted by the present batch experiments through the solution of the underlying optimal control problems by means of stable and accurate discretisation techniques. It was found that an initial glutamate to glycerol ratio of 0.25 was optimal and was independent of the initial biomass concentration. The glycerol conversion efficiency was found to depend on initial biomass concentration and its computational peak is 64.4%. By optimising a 30-day industrially relevant batch process, the hydrogen productivity was improved to be 37.7 mL g biomass−1 h−1 and the glycerol conversion efficiency was maintained at 58%. The models can then be applied as the connection to transfer biohydrogen production from laboratory scale into industrial scale.

The focus of this paper is the description and numerical validation of a computational method where stress analysis can be performed directly from computer-aided design data without mesh generation. The clear benefit of the approach is that no mesh needs to be generated prior to running the analysis. This is achieved by utilising the isogeometric concept whereby computer-aided design data are used to construct not only the geometry discretisation but also the displacement and traction approximations. In this manner, significant savings can be made in the engineering design and analysis process. This paper also demonstrates that, compared with a standard boundary-element method implementation using quadratic Lagrangian shape functions, superior accuracy is achieved using the present approach for the same number of degrees of freedom. It further illustrates practical applications of the method, comparing against results obtained with a standard boundary-element method and finite-element method for verification. In addition, a propeller is analysed as a sample to show the ability of the present method to handle complex three-dimensional geometries.

[Departement_IRSTEA]Eaux [TR1_IRSTEA]ARCEAU; Distributed hydrological models are valuable tools that can be used to support water management in catchments. However, the complexity of management issues, the variety of modelling objectives, and the variable availability of data require a flexible way to customize models and adapt them to each individual problem. Environmental modelling frameworks offer such flexibility; they are designed to build and run integrated models on the basis of reusable and exchangeable components. This paper presents the LIQUID® framework, developed by Hydrowide since 2005. The purpose of developing LIQUID® was to provide both easier integration of hydrological processes and preservation of their characteristic temporal and spatial scales. It suits a wide range of applications, both in terms of spatial scales and of process conceptualisations. LIQUID® is able to synchronize different time steps, to handle irregular geometries, and to simulate complex connections between components, in particular involving feedback. The paper presents the concepts of LIQUID® and the technical choices made to meet the above requirements, with focuses on the simulation run system and on the spatial discretization of process components. The use of the framework is illustrated by five application cases associated with contrasted spatial and temporal scales.