Probability density estimation with tunable kernels using orthogonal forward regression

A generalized or tunable-kernel model is proposed for probability density function estimation based on an orthogonal forward regression procedure. Each stage of the density estimation process determines a tunable kernel, namely, its center vector and diagonal covariance matrix, by minimizing a leave-one-out test criterion. The kernel mixing weights of the constructed sparse density estimate are finally updated using the multiplicative nonnegative quadratic programming algorithm to ensure the nonnegative and unity constraints, and this weight-updating process additionally has the desired ability to further reduce the model size. The proposed tunable-kernel model has advantages, in terms of model generalization capability and model sparsity, over the standard fixed-kernel model that restricts kernel centers to the training data points and employs a single common kernel variance for every kernel. On the other hand, it does not optimize all the model parameters together and thus avoids the problems of high-dimensional ill-conditioned nonlinear optimization associated with the conventional finite mixture model. Several examples are included to demonstrate the ability of the proposed novel tunable-kernel model to effectively construct a very compact density estimate accurately.

System identification using partitioned least squares

A novel partitioned least squares (PLS) algorithm is presented, in which estimates from several simple system models are combined by means of a Bayesian methodology of pooling partial knowledge. The method has the added advantage that, when the simple models are of a similar structure, it lends itself directly to parallel processing procedures, thereby speeding up the entire parameter estimation process by several factors.

Complex-valued B-spline neural networks for modeling and inverting Hammerstein systems

Many communication signal processing applications involve modelling and inverting complex-valued (CV) Hammerstein systems. We develops a new CV B-spline neural network approach for efficient identification of the CV Hammerstein system and effective inversion of the estimated CV Hammerstein model. Specifically, the CV nonlinear static function in the Hammerstein system is represented using the tensor product from two univariate B-spline neural networks. An efficient alternating least squares estimation method is adopted for identifying the CV linear dynamic model’s coefficients and the CV B-spline neural network’s weights, which yields the closed-form solutions for both the linear dynamic model’s coefficients and the B-spline neural network’s weights, and this estimation process is guaranteed to converge very fast to a unique minimum solution. Furthermore, an accurate inversion of the CV Hammerstein system can readily be obtained using the estimated model. In particular, the inversion of the CV nonlinear static function in the Hammerstein system can be calculated effectively using a Gaussian-Newton algorithm, which naturally incorporates the efficient De Boor algorithm with both the B-spline curve and first order derivative recursions. The effectiveness of our approach is demonstrated using the application to equalisation of Hammerstein channels.

An empirical study of process-related attributes in segmented software cost-estimation relationships

Parametric software effort estimation models consisting on a single mathematical relationship suffer from poor adjustment and predictive characteristics in cases in which the historical database considered contains data coming from projects of a heterogeneous nature. The segmentation of the input domain according to clusters obtained from the database of historical projects serves as a tool for more realistic models that use several local estimation relationships. Nonetheless, it may be hypothesized that using clustering algorithms without previous consideration of the influence of well-known project attributes misses the opportunity to obtain more realistic segments. In this paper, we describe the results of an empirical study using the ISBSG-8 database and the EM clustering algorithm that studies the influence of the consideration of two process-related attributes as drivers of the clustering process: the use of engineering methodologies and the use of CASE tools. The results provide evidence that such consideration conditions significantly the final model obtained, even though the resulting predictive quality is of a similar magnitude.

Estimation of Gaussian process regression model using probability distance measures

A new class of parameter estimation algorithms is introduced for Gaussian process regression (GPR) models. It is shown that the integration of the GPR model with probability distance measures of (i) the integrated square error and (ii) Kullback–Leibler (K–L) divergence are analytically tractable. An efficient coordinate descent algorithm is proposed to iteratively estimate the kernel width using golden section search which includes a fast gradient descent algorithm as an inner loop to estimate the noise variance. Numerical examples are included to demonstrate the effectiveness of the new identification approaches.

Uncertainty assessment of a process-based integrated catchment model of phosphorus

Despite the many models developed for phosphorus concentration prediction at differing spatial and temporal scales, there has been little effort to quantify uncertainty in their predictions. Model prediction uncertainty quantification is desirable, for informed decision-making in river-systems management. An uncertainty analysis of the process-based model, integrated catchment model of phosphorus (INCA-P), within the generalised likelihood uncertainty estimation (GLUE) framework is presented. The framework is applied to the Lugg catchment (1,077 km2), a River Wye tributary, on the England–Wales border. Daily discharge and monthly phosphorus (total reactive and total), for a limited number of reaches, are used to initially assess uncertainty and sensitivity of 44 model parameters, identified as being most important for discharge and phosphorus predictions. This study demonstrates that parameter homogeneity assumptions (spatial heterogeneity is treated as land use type fractional areas) can achieve higher model fits, than a previous expertly calibrated parameter set. The model is capable of reproducing the hydrology, but a threshold Nash-Sutcliffe co-efficient of determination (E or R 2) of 0.3 is not achieved when simulating observed total phosphorus (TP) data in the upland reaches or total reactive phosphorus (TRP) in any reach. Despite this, the model reproduces the general dynamics of TP and TRP, in point source dominated lower reaches. This paper discusses why this application of INCA-P fails to find any parameter sets, which simultaneously describe all observed data acceptably. The discussion focuses on uncertainty of readily available input data, and whether such process-based models should be used when there isn’t sufficient data to support the many parameters.

Vegetation dynamics in abandoned crop fields on a Mediterranean island: Development of succession model and estimation of disturbance thresholds

This case study on the Sifnos island, Greece, assesses the main factors controlling vegetation succession following crop abandonment and describes the vegetation dynamics of maquis and phrygana formations in relation to alternative theories of secondary succession. Field survey data were collected and analysed at community as well as species level. The results show that vegetation succession on abandoned crop fields is determined by the combined effects of grazing intensity, soil and geological characteristics and time. The analysis determines the quantitative grazing thresholds that modify the successional pathway. Light grazing leads to dominance by maquis vegetation while overgrazing leads to phryganic vegetation. The proposed model shows that vegetation succession following crop abandonment is a complex multi-factor process where the final or the stable stage of the process is not predefined but depends on the factors affecting succession. An example of the use of succession models and disturbance thresholds as a policy assessment tool is presented by evaluating the likely vegetation impacts of the recent reform of the Common Agricultural Policy on Sifnos island over a 20-30-year time horizon. (c) 2006 Elsevier B.V. All rights reserved.

Probability density function estimation using orthogonal forward regression

Using the classical Parzen window estimate as the target function, the kernel density estimation is formulated as a regression problem and the orthogonal forward regression technique is adopted to construct sparse kernel density estimates. The proposed algorithm incrementally minimises a leave-one-out test error score to select a sparse kernel model, and a local regularisation method is incorporated into the density construction process to further enforce sparsity. The kernel weights are finally updated using the multiplicative nonnegative quadratic programming algorithm, which has the ability to reduce the model size further. Except for the kernel width, the proposed algorithm has no other parameters that need tuning, and the user is not required to specify any additional criterion to terminate the density construction procedure. Two examples are used to demonstrate the ability of this regression-based approach to effectively construct a sparse kernel density estimate with comparable accuracy to that of the full-sample optimised Parzen window density estimate.

An orthogonal forward regression technique for sparse kernel density estimation

Using the classical Parzen window (PW) estimate as the desired response, the kernel density estimation is formulated as a regression problem and the orthogonal forward regression technique is adopted to construct sparse kernel density (SKD) estimates. The proposed algorithm incrementally minimises a leave-one-out test score to select a sparse kernel model, and a local regularisation method is incorporated into the density construction process to further enforce sparsity. The kernel weights of the selected sparse model are finally updated using the multiplicative nonnegative quadratic programming algorithm, which ensures the nonnegative and unity constraints for the kernel weights and has the desired ability to reduce the model size further. Except for the kernel width, the proposed method has no other parameters that need tuning, and the user is not required to specify any additional criterion to terminate the density construction procedure. Several examples demonstrate the ability of this simple regression-based approach to effectively construct a SKID estimate with comparable accuracy to that of the full-sample optimised PW density estimate. (c) 2007 Elsevier B.V. All rights reserved.

Application of a novel optimal control algorithm to a benchmark fed-batch fermentation process

A novel algorithm for solving nonlinear discrete time optimal control problems with model-reality differences is presented. The technique uses Dynamic Integrated System Optimisation and Parameter Estimation (DISOPE) which has been designed to achieve the correct optimal solution in spite of deficiencies in the mathematical model employed in the optimisation procedure. A method based on Broyden's ideas is used for approximating some derivative trajectories required. Ways for handling con straints on both manipulated and state variables are described. Further, a method for coping with batch-to- batch dynamic variations in the process, which are common in practice, is introduced. It is shown that the iterative procedure associated with the algorithm naturally suits applications to batch processes. The algorithm is success fully applied to a benchmark problem consisting of the input profile optimisation of a fed-batch fermentation process.

Dynamic data reconciliation for sequential modular simulators: application to a mixing process

This paper describes a method for dynamic data reconciliation of nonlinear systems that are simulated using the sequential modular approach, and where individual modules are represented by a class of differential algebraic equations. The estimation technique consists of a bank of extended Kalman filters that are integrated with the modules. The paper reports a study based on experimental data obtained from a pilot scale mixing process.

New developments in batch process input profile optimisation

A novel algorithm for solving nonlinear discrete time optimal control problems with model-reality differences is presented. The technique uses dynamic integrated system optimisation and parameter estimation (DISOPE) which achieves the correct optimal solution in spite of deficiencies in the mathematical model employed in the optimisation procedure. A new method for approximating some Jacobian trajectories required by the algorithm is introduced. It is shown that the iterative procedure associated with the algorithm naturally suits applications to batch chemical processes.

Hierarchical control and optimisation of the Tennessee Eastman process

Based on integrated system optimisation and parameter estimation a method is described for on-line steady state optimisation which compensates for model-plant mismatch and solves a non-linear optimisation problem by iterating on a linear - quadratic representation. The method requires real process derivatives which are estimated using a dynamic identification technique. The utility of the method is demonstrated using a simulation of the Tennessee Eastman benchmark chemical process.

State estimation using the particle filter with mode tracking

A particle filter is a data assimilation scheme that employs a fully nonlinear, non-Gaussian analysis step. Unfortunately as the size of the state grows the number of ensemble members required for the particle filter to converge to the true solution increases exponentially. To overcome this Vaswani [Vaswani N. 2008. IEEE Trans Signal Process 56:4583–97] proposed a new method known as mode tracking to improve the efficiency of the particle filter. When mode tracking, the state is split into two subspaces. One subspace is forecast using the particle filter, the other is treated so that its values are set equal to the mode of the marginal pdf. There are many ways to split the state. One hypothesis is that the best results should be obtained from the particle filter with mode tracking when we mode track the maximum number of unimodal dimensions. The aim of this paper is to test this hypothesis using the three dimensional stochastic Lorenz equations with direct observations. It is found that mode tracking the maximum number of unimodal dimensions does not always provide the best result. The best choice of states to mode track depends on the number of particles used and the accuracy and frequency of the observations.

Sea surface temperature from a geostationary satellite by optimal estimation

Optimal estimation (OE) is applied as a technique for retrieving sea surface temperature (SST) from thermal imagery obtained by the Spinning Enhanced Visible and Infra-Red Imager (SEVIRI) on Meteosat 9. OE requires simulation of observations as part of the retrieval process, and this is done here using numerical weather prediction fields and a fast radiative transfer model. Bias correction of the simulated brightness temperatures (BTs) is found to be a necessary step before retrieval, and is achieved by filtered averaging of simulations minus observations over a time period of 20 days and spatial scale of 2.5° in latitude and longitude. Throughout this study, BT observations are clear-sky averages over cells of size 0.5° in latitude and longitude. Results for the OE SST are compared to results using a traditional non-linear retrieval algorithm (“NLSST”), both validated against a set of 30108 night-time matches with drifting buoy observations. For the OE SST the mean difference with respect to drifter SSTs is − 0.01 K and the standard deviation is 0.47 K, compared to − 0.38 K and 0.70 K respectively for the NLSST algorithm. Perhaps more importantly, systematic biases in NLSST with respect to geographical location, atmospheric water vapour and satellite zenith angle are greatly reduced for the OE SST. However, the OE SST is calculated to have a lower sensitivity of retrieved SST to true SST variations than the NLSST. This feature would be a disadvantage for observing SST fronts and diurnal variability, and raises questions as to how best to exploit OE techniques at SEVIRI's full spatial resolution.