A novel continuous forward algorithm for RBF neural modelling

A continuous forward algorithm (CFA) is proposed for nonlinear modelling and identification using radial basis function (RBF) neural networks. The problem considered here is simultaneous network construction and parameter optimization, well-known to be a mixed integer hard one. The proposed algorithm performs these two tasks within an integrated analytic framework, and offers two important advantages. First, the model performance can be significantly improved through continuous parameter optimization. Secondly, the neural representation can be built without generating and storing all candidate regressors, leading to significantly reduced memory usage and computational complexity. Computational complexity analysis and simulation results confirm the effectiveness.

Application of static charge transfer within an ionic-liquid force field and its effect on structure and dynamics

The effects of linear scaling of the atomic charges of a reference potential on the structure, dynamics, and energetics of the ionic liquid 1,3-dimethylimidazolium chloride are investigated. Diffusion coefficients that span over four orders of magnitude are observed between the original model and a scaled model in which the ionic charges are +/- 0.5 e. While the three-dimensional structure of the liquid is less affected, the partial radial distribution functions change markedly-with the positive result that for ionic charges of +/- 0.7 e, an excellent agreement is observed with ab initio molecular dynamics data. Cohesive energy densities calculated from these partial-charge models are also in better agreement with those calculated from the ab initio data. We postulate that ionic-liquid models in which the ionic charges are assumed to be +/- 1 e overestimate the intermolecular attractions between ions, which results in overstructuring, slow dynamics, and increased cohesive energy densities. The use of scaled-charge sets may be of benefit in the simulation of these systems-especially when looking at properties beyond liquid structure-thus providing on alternative to computationally expensive polarisable force fields.

Characterization of a high-density, direct-current reflex discharge plasma source operating in Ar and N-2

The characterization of a direct current, low-pressure, and high-density reflex discharge plasma source operating in argon and in nitrogen, over a range of pressures 1.0-10(-2) mbar, discharge currents 20-200 mA, and magnetic fields 0-120 G, and its parametric characterization is presented. Both external parameters, such as the breakdown potential and the discharge voltage-current characteristic, and internal parameters, like the charge carrier's temperature and density, plasma potential, floating potential, and electron energy distribution function, were measured. The electron energy distribution functions are bi-Maxwellian, but some structure is observed in these functions in nitrogen plasmas. There is experimental evidence for the existence of three groups of electrons within this reflex discharge plasma. Due to the enhanced hollow cathode effect by the magnetic trapping of electrons, the density of the cold group of electrons is as high as 10(18) m(-3), and the temperature is as low as a few tenths of an electron volt. The bulk plasma density scales with the dissipated power. Another important feature of this reflex plasma source is its high degree of uniformity, while the discharge bulk region is free of electric field. (C) 2002 American Institute of Physics.

High-density and low electron temperature direct current reflex plasma source

A new type of direct current, high-density, and low electron temperature reflex plasma source, obtained as a hybrid between a modified hollow-cathode discharge and a Penning ionization gauge discharge is presented. The plasma source was tested in argon, nitrogen, and oxygen over a range pressure of 1.0-10(-3) mbar, discharge currents 20-200 mA, and magnetic field 0-120 Gauss. Both external parameters, such as breakdown potential and the discharge voltage-current characteristic, and its internal parameters, like the electron energy distribution function, electron and ion densities, and electron temperature, were measured. Due to the enhanced hollow-cathode effect by the magnetic trapping of electrons, the density of the bulk plasma is as high as 10(18) m(-3), and the electron temperature is as low as a few tenths of electron volts. The plasma density scales with the dissipated power. Another important feature of this reflex plasma source is its high degree of uniformity, while the discharge bulk region is free of an electric field. (C) 2004 American Institute of Physics.

PLS modelling and fault detection on the Tennessee Eastman benchmark

This paper describes the application of multivariate regression techniques to the Tennessee Eastman benchmark process for modelling and fault detection. Two methods are applied : linear partial least squares, and a nonlinear variant of this procedure using a radial basis function inner relation. The performance of the RBF networks is enhanced through the use of a recently developed training algorithm which uses quasi-Newton optimization to ensure an efficient and parsimonious network; details of this algorithm can be found in this paper. The PLS and PLS/RBF methods are then used to create on-line inferential models of delayed process measurements. As these measurements relate to the final product composition, these models suggest that on-line statistical quality control analysis should be possible for this plant. The generation of `soft sensors' for these measurements has the further effect of introducing a redundant element into the system, redundancy which can then be used to generate a fault detection and isolation scheme for these sensors. This is achieved by arranging the sensors and models in a manner comparable to the dedicated estimator scheme of Clarke et al. 1975, IEEE Trans. Pero. Elect. Sys., AES-14R, 465-473. The effectiveness of this scheme is demonstrated on a series of simulated sensor and process faults, with full detection and isolation shown to be possible for sensor malfunctions, and detection feasible in the case of process faults. Suggestions for enhancing the diagnostic capacity in the latter case are covered towards the end of the paper.

Neural network based control for synchronous generators

In this paper, a Radial Basis Function neural network based AVR is proposed. A control strategy which generates local linear models from a global neural model on-line is used to derive controller feedback gains based on the Generalised Minimum Variance technique. Testing is carried out on a micromachine system which enables evaluation of practical implementation of the scheme. Constraints imposed by gathering training data, computational load, and memory requirements for the training algorithm are addressed.

Diagnostic-based modeling on a micro-scale atmospheric-pressure plasma jet

Diagnostic-based modeling (DBM) actively combines complementary advantages of numerical plasma simulations and relatively simple optical emission spectroscopy (OES). DBM is applied to determine spatial absolute atomic oxygen ground-state density profiles in a micro atmospheric-pressure plasma jet operated in He–O2. A 1D fluid model with semi-kinetic treatment of the electrons yields detailed information on the electron dynamics and the corresponding spatio-temporal electron energy distribution function. Benchmarking this time- and space-resolved simulation with phase-resolved OES (PROES) allows subsequent derivation of effective excitation rates as the basis for DBM. The population dynamics of the upper O(3p3P) oxygen state (? = 844 nm) is governed by direct electron impact excitation, dissociative excitation, radiation losses, and collisional induced quenching. Absolute values for atomic oxygen densities are obtained through tracer comparison with the upper Ar(2p1) state (? = 750.4 nm). The resulting spatial profile for the absolute atomic oxygen density shows an excellent quantitative agreement to a density profile obtained by two-photon absorption laser-induced fluorescence spectroscopy.

A fast multi-output RBF neural network construction method

This paper investigates the center selection of multi-output radial basis function (RBF) networks, and a multi-output fast recursive algorithm (MFRA) is proposed. This method can not only reveal the significance of each candidate center based on the reduction in the trace of the error covariance matrix, but also can estimate the network weights simultaneously using a back substitution approach. The main contribution is that the center selection procedure and the weight estimation are performed within a well-defined regression context, leading to a significantly reduced computational complexity. The efficiency of the algorithm is confirmed by a computational complexity analysis, and simulation results demonstrate its effectiveness. (C) 2010 Elsevier B.V. All rights reserved.

Modeling Multivariate Interest Rates Using Time-Varying Copulas and Reducible Nonlinear Stochastic Differential Equations

We propose a new approach for modeling nonlinear multivariate interest rate processes based on time-varying copulas and reducible stochastic differential equations (SDEs). In the modeling of the marginal processes, we consider a class of nonlinear SDEs that are reducible to Ornstein--Uhlenbeck (OU) process or Cox, Ingersoll, and Ross (1985) (CIR) process. The reducibility is achieved via a nonlinear transformation function. The main advantage of this approach is that these SDEs can account for nonlinear features, observed in short-term interest rate series, while at the same time leading to exact discretization and closed-form likelihood functions. Although a rich set of specifications may be entertained, our exposition focuses on a couple of nonlinear constant elasticity volatility (CEV) processes, denoted as OU-CEV and CIR-CEV, respectively. These two processes encompass a number of existing models that have closed-form likelihood functions. The transition density, the conditional distribution function, and the steady-state density function are derived in closed form as well as the conditional and unconditional moments for both processes. In order to obtain a more flexible functional form over time, we allow the transformation function to be time varying. Results from our study of U.S. and UK short-term interest rates suggest that the new models outperform existing parametric models with closed-form likelihood functions. We also find the time-varying effects in the transformation functions statistically significant. To examine the joint behavior of interest rate series, we propose flexible nonlinear multivariate models by joining univariate nonlinear processes via appropriate copulas. We study the conditional dependence structure of the two rates using Patton (2006a) time-varying symmetrized Joe--Clayton copula. We find evidence of asymmetric dependence between the two rates, and that the level of dependence is positively related to the level of the two rates. (JEL: C13, C32, G12) Copyright The Author 2010. Published by Oxford University Press. All rights reserved. For permissions, please e-mail: journals.permissions@oxfordjournals.org, Oxford University Press.

Spatial evolution of a q-Gaussian laser beam in relativistic plasma

In a recent experimental study, the beam intensity profile of the Vulcan petawatt laser beam was measured; it was found that only 20% of the energy was contained within the full width at half maximum of 6.9 mu m and 50% within 16 mu m, suggesting a long-tailed non-Gaussian transverse beam profile. A q-Gaussian distribution function was suggested therein to reproduce this behavior. The spatial beam profile dynamics of a q-Gaussian laser beam propagating in relativistic plasma is investigated in this article. A non-paraxial theory is employed, taking into account nonlinearity via the relativistic decrease of the plasma frequency. We have studied analytically and numerically the dynamics of a relativistically guided beam and its dependence on the q-parameter. Numerical simulation results are shown to trace the dependence of the focusing length on the q-Gaussian profile.

Fokker-Planck equation for a test-particle weakly coupled to a magnetized plasma

We consider the derivation of a kinetic equation for a charged test particle weakly interacting with an electrostatic plasma in thermal equilibrium, subject to a uniform external magnetic field. The Liouville equation leads to a generalized master equation to second order in the `weak' interaction; a Fokker-Planck-type equation then follows as a `Markovian' approximation. It is shown that such an equation does not preserve the positivity of the distribution function f(x,v;t). By applying techniques developed in the theory of open systems, a correct Fokker-Planck equation is derived. Explicit expressions for the diffusion and drift coefficients, depending on the magnetic field, are obtained.

Locally regularised two-stage learning algorithm for RBF network centre selection

Nonlinear models constructed from radial basis function (RBF) networks can easily be over-fitted due to the noise on the data. While information criteria, such as the final prediction error (FPE), can provide a trade-off between training error and network complexity, the tunable parameters that penalise a large size of network model are hard to determine and are usually network dependent. This article introduces a new locally regularised, two-stage stepwise construction algorithm for RBF networks. The main objective is to produce a parsomous network that generalises well over unseen data. This is achieved by utilising Bayesian learning within a two-stage stepwise construction procedure to penalise centres that are mainly interpreted by the noise.

Application of Nonlinear PCA for Fault Detection in Polymer Extrusion Processes

This paper describes the application of an improved nonlinear principal component analysis (PCA) to the detection of faults in polymer extrusion processes. Since the processes are complex in nature and nonlinear relationships exist between the recorded variables, an improved nonlinear PCA, which incorporates the radial basis function (RBF) networks and principal curves, is proposed. This algorithm comprises two stages. The first stage involves the use of the serial principal curve to obtain the nonlinear scores and approximated data. The second stage is to construct two RBF networks using a fast recursive algorithm to solve the topology problem in traditional nonlinear PCA. The benefits of this improvement are demonstrated in the practical application to a polymer extrusion process.

Two-stage RBF network construction based on particle swarm optimization

The conventional radial basis function (RBF) network optimization methods, such as orthogonal least squares or the two-stage selection, can produce a sparse network with satisfactory generalization capability. However, the RBF width, as a nonlinear parameter in the network, is not easy to determine. In the aforementioned methods, the width is always pre-determined, either by trial-and-error, or generated randomly. Furthermore, all hidden nodes share the same RBF width. This will inevitably reduce the network performance, and more RBF centres may then be needed to meet a desired modelling specification. In this paper we investigate a new two-stage construction algorithm for RBF networks. It utilizes the particle swarm optimization method to search for the optimal RBF centres and their associated widths. Although the new method needs more computation than conventional approaches, it can greatly reduce the model size and improve model generalization performance. The effectiveness of the proposed technique is confirmed by two numerical simulation examples.

Nonrigid Structure-From-Motion From 2-D Images Using Markov Chain Monte Carlo

In this paper we present a new method for simultaneously determining three dimensional (3-D) shape and motion of a non-rigid object from uncalibrated two dimensional (2- D) images without assuming the distribution characteristics. A non-rigid motion can be treated as a combination of a rigid rotation and a non-rigid deformation. To seek accurate recovery of deformable structures, we estimate the probability distribution function of the corresponding features through random sampling, incorporating an established probabilistic model. The fitting between the observation and the projection of the estimated 3-D structure will be evaluated using a Markov chain Monte Carlo based expectation maximisation algorithm. Applications of the proposed method to both synthetic and real image sequences are demonstrated with promising results.