Multivariate Parametric Models Research Articles

Robust inference by influence functions

We first review briefly some basic approaches to robust inference and discuss the role and the place of some key concepts (influence function, breakdown point, robustness versus efficiency, etc.). We then discuss in some detail recent results on robust testing in general multivariate parametric models. Recent applications include inference in logistic regression and testing for non-nested hypotheses.

A Stochastic Minimum Distance Test for Multivariate Parametric Models

Stochastic procedures are randomized statistical procedures which are functions of the observed sample and of one or more artificially constructed auxiliary samples. As the size of the auxiliary samples increases, a stochastic procedure becomes nearly nonrandomized. The stochastic test of this paper arises as a numerically feasible approximation to a natural minimum distance goodness-of-fit test for multivariate parametric models. The distance being minimized here is the half-space metric for probabilities on a Euclidean space. It is shown that the various approximations used in constructing the stochastic test and its critical values do not detract from its first-order asymptotic performance.

Parameter Adaptive Modelling of a 500 Megawatt Drum Boiler

Measurements of the coal feed, feedwater flow, drum level, steam flow, turbine steam valve (TSV) pressure error and generator power were monitored on a 500 MW(e) drum boiler during full power operation under closed loop control. The fluctuations in these measurements about their mean value are analysed using multivariate time series techniques and parametric models to Identify the dynamics of the drum boiler. Recursive parameter estimation techniques and identlflability problems In determining the ‘optimum’ order autoregressive moving average (ARMA) models of the feedforward (process) and feedback (recycle/control) paths of distributed closed loop systems are discussed. The advantages of using cross-correlation techniques for systems involving unknown feedback structure are illustrated. Parametric models of the variable affecting drum level are then identified using a recursive parameter estimation methodology which Incorporates a covariance resetting procedure.

Multivariate autoregressive representation of seismic P-wave signals with application to short-period discrimination

abstract It is shown that seismic P-wave vector signals as recorded by selected NORSAR subarrays can be described by multivariate parametric models of autoregressive type. These are models having the form X ¯ ( t ) − A 1 X ¯ ( t − 1 ) − ... − A p X ¯ ( t − p ) = W ¯ ( t ) Where X¯(t) (t) is the digitized short-period vector time series defined by the P-wave signal and W¯(t) (t) is a white noise vector time series. The multivariate autoregressive analysis is undertaken for 83 nuclear explosions and 72 earthquakes from Eurasia. For each event a separate analysis of the main signal and of the coda has been carried through. It is found that in most cases a reasonable statistical fit is obtained using a low-order autoregressive model. The autoregressive parameters characterize the spectral density matrix of the P-wave signal and therefore form a convenient basis for constructing short-period discriminants between earthquakes and explosions. Based on the classification results for our data base of Eurasian events, we find that the multivariate autoregressive parameters have a substantially larger discrimination potential than the short-period parameters suggested earlier in the literature. In fact our results indicate that, based on autoregressive parameters, it may now be possible to construct purely short-period discriminants which are comparable, if not superior, to the mb:Ms criterion.