The least squares algorithm, parametric system identification and bounded noise

The least squares parametric system identification algorithm is analyzed assuming that the noise is a bounded signal. A bound on the worst-case parameter estimation error is derived. This bound shows that the worst-case parameter estimation error decreases to zero as the bound on the noise is decreased to zero.     

H∞ identification of “soft” uncertainty models

The paper investigates the problem of identifying uncertainty models of causal, SISO, LTI, discrete-time, BIBO stable, unknown systems, using frequency domain measurements corrupted by Gaussian noise of known covariance. Additive uncertainty models are looked for, consisting of a nominal model and an additive dynamic perturbation accounting for the modeling error. The nominal model is chosen within a class of affinely parametrized models with transfer function of given (possibly low) order. An estimate of the parameters minimizing the H_∞ modeling error is obtained by minimizing an upper bound of the worst-case (with respect to the modeling error) second moment of the estimation error. Then, a bound in the frequency domain guaranteeing to include, with probability α, the frequency response error between the estimated nominal model and the unknown system is derived.     

Distributed and Recursive Parameter Estimation in Parametrized Linear State-Space Models

We consider a network of sensors deployed to sense a spatio-temporal field and infer parameters of interest about the field. We are interested in the case where each sensor's observation sequence is modeled as a state-space process that is perturbed by random noise, and the models across sensors are parametrized by the same parameter vector. The sensors collaborate to estimate this parameter from their measurements, and to this end we propose a distributed and recursive estimation algorithm, which we refer to as the incremental recursive prediction error algorithm. This algorithm has the distributed property of incremental gradient algorithms and the on-line property of recursive prediction error algorithms.      

Modified adaptive robust control system design

The robust control design problem is studied for a class of uncertain dynamical systems. The uncertainty of the system is time varying. The only assumption on the uncertainty is that it is bounded. No statistical property of the uncertainty is ever assumed and utilized. The robust control does not need the a priori estimation on the bound of uncertainty. An adaptive algorithm for the on-line estimation of the bound is constructed and the robust control is only dependent on this estimation. The adaptive algorithm is a modification of previous work by Corless and Leitmann (1983). The advantages of this new algorithm include a constant control design parameter (which should have been time varying previously) and the applicability to linear systems with mismatched uncertainty and measurement noise.     

Pi-coefficient analysis of update algorithms for adaptive systems

This paper finds the appropriate pi-coefficients for a parameter estimation adaptive system and uses them to analyze the stability of two estimation algorithms. The estimation error dynamics of the system are modeled by a linear time-invariant subsystem and a nonlinear time-varying update law in a feedback loop. Then the so-called max-p problems are formulated and solved to obtain the pi-coefficients for the linear subsystem and nonlinear update low. For the investigated system, the quantitative results show that the least-squares update algorithm has larger stability range than that of the gradient algorithm, and the σ-modification scheme gives larger stability ranges for both algorithms.     

Adaptive λ-tracking for a class of infinite-dimensional systems

For a class of high-gain stabilizable multivariable linear infinite-dimensional systems we present an adaptive control law which achieves approximate asymptotic tracking in the sense that the tracking error tends asymptotically to a ball centred at 0 and of arbitrary prescribed radius λ>0. This control strategy, called λ-tracking, combines proportional error feedback with a simple nonlinear adaptation of the feedback gain. It does not involve any parameter estimation algorithms, nor is it based on the internal model principle. The class of reference signals is W^1,∞, the Sobolev space of absolutely continuous functions which are bounded and have essentially bounded derivative. The control strategy is robust with respect to output measurement noise in W^1,∞ and bounded input disturbances. We apply our results to retarded systems and integrodifferential systems.     

Reduced-basis output bounds for approximately parametrized elliptic coercive partial differential equations

We present a technique for the rapid and reliable prediction of linear-functional outputs of elliptic coercive partial differential equations with (approximately) affine parameter dependence. The essential components are (i) (provably) rapidly convergent global reduced-basis approximations – Galerkin projection onto a space W_N spanned by solutions of the governing partial differential equation at N selected points in parameter space; (ii) a posteriori error estimation – relaxations of the error-residual equation that provide inexpensive yet sharp bounds for the error in the outputs of interest; and (iii) off-line/on-line computational procedures – methods which decouple the generation and projection stages of the approximation process. The operation count for the on-line stage – in which, given a new parameter value, we calculate the output of interest and associated error bound – depends only on N, typically very small, and the (approximate) parametric complexity of the problem; the method is thus ideally suited for the repeated and rapid evaluations required in the context of parameter estimation, design, optimization, and real-time control.In our earlier work, we develop a rigorous a posteriori error bound framework for the case in which the parametrization of the partial differential equation is exact; in this paper, we address the situation in which our mathematical model is not complete. In particular, we permit error in the data that define our partial differential operator: this error may be introduced, for example, by imperfect specification, measurement, calculation, or parametric expansion of a coefficient function. We develop both accurate predictions for the outputs of interest and associated rigorous a posteriori error bounds; and the latter incorporate both numerical discretization and model truncation effects. Numerical results are presented for a particular instantiation in which the model error originates in the (approximately) prescribed velocity field associated with a three-dimensional convection-diffusion problem.     

Quadratic optimal neural fuzzy control for synchronization of uncertain chaotic systems

This paper presents a novel quadratic optimal neural fuzzy control for synchronization of uncertain chaotic systems via H^∞ approach. In the proposed algorithm, a self-constructing neural fuzzy network (SCNFN) is developed with both structure and parameter learning phases, so that the number of fuzzy rules and network parameters can be adaptively determined. Based on the SCNFN, an uncertainty observer is first introduced to watch compound system uncertainties. Subsequently, an optimal NFN-based controller is designed to overcome the effects of unstructured uncertainty and approximation error by integrating the NFN identifier, linear optimal control and H^∞ approach as a whole. The adaptive tuning laws of network parameters are derived in the sense of quadratic stability technique and Lyapunov synthesis approach to ensure the network convergence and H^∞ synchronization performance. The merits of the proposed control scheme are not only that the conservative estimation of NFN approximation error bound is avoided but also that a suitable-sized neural structure is found to sufficiently approximate the system uncertainties. Simulation results are provided to verify the effectiveness and robustness of the proposed control method.     

ON SYSTEM IDENTIFICATION WITH AND WITHOUT CERTAINTY

Abstract

Information-theoretic concepts are utilized to develop a procedure for identifying a parameter of a stochastic linear discrete time dynamic scalar system based on noisy linear measurements of the system's state. After various simplifying approximations, the derived error entropy identification algorithm reduces to an on-line adaptive identification algorithm that is similar in many respects to well-established identification techniques. Conditions under which the developed on-line adaptive algorithm identifies the system with certainty are presented. Using an error entropy estimation lower bound, which is independent of any estimation procedure, conditions for which identification cannot be made with certainty are also presented. Examples involving non-Gaussian statistics are used to illustrate the efficiency of the error entropy adaptive identification algorithm as well as to compare it with several other identification procedures.     

时变参数遗忘梯度估计算法的收敛性

提出了时变随机系统的遗忘梯度辨识算法,并运用随机过程理论研究了算法的收敛 性.分析表明,遗忘梯度算法的性能类似于遗忘因子最小二乘法,可以跟踪时变参数,但计算量 要小得多,且数据的平稳性可以减小参数估计误差上界和提高辨识精度.阐述了最佳遗忘因子 的选择方法,以获得最小参数估计上界.对于确定性时不变系统,遗忘梯度算法是指数速度收 敛的;对于时变或时不变随机系统,遗忘梯度算法的参数估计误差一致有上界.     

Intelligent Control of Discrete Linear Systems Based on a Supervised Adaptive Multiestimation Scheme

A pole-placement based adaptive controller synthesised from a multiestimation scheme is designed for linear plants. A higher level switching structure between the various estimation schemes is used to supervise the reparameterisation of the adaptive controller in real time. The basic usefulness of the proposed scheme is to improve the transient response so that the closed-loop stability is guaranteed. The switching process is subject to a minimum dwelling or residence time within which the supervisor is not allowed to switch between the multiple estimation schemes. The high level supervision is based on the multiestimation identification scheme. The residence time condition guarantees the closed-loop stability. The above higher level switching structure is on-line supervised by a closed-loop tracking error based algorithm. This second supervision on-line tunes the free design parameters which appear as time varying weights in the loss function of the above switching structure. Thus, the closed-loop behaviour, compared to the constant parameter case one, is improved when the design parameter is not tightly initialised. Both supervisors are hierarchically organised in the sense that they act on the system at different rates. Furthermore, a projection algorithm has been considered in the estimation scheme in order to include a possible a priori knowledge of the estimates parameter vector value in the estimation algorithm.     

Worst case identification of continuous time systems via interpolation

We consider a worst case robust control oriented identification problem recently studied by several authors. This problem is one of identification in the continuous time setting. We give a more general formulation of this problem. The available a priori information in this paper consists of a lower bound on the relative stability of the plant, a frequency dependent upper bound on a certain gain associated with the plant, and an upper bound on the noise level. The available experimental information consists of a finite number of noisy plant point frequency response samples. The objective is to identify, from the given a priori and experimental information, an uncertain model that includes a stable nominal plant model and a bound on the modeling error measured in norm. Our main contributions include both a new identification algorithm and several new ‘explicit’ lower and upper bounds on the identification error. The proposed algorithm belongs to the class of ‘interpolatory algorithms’ which are known to possess a desirable optimality property under a certain criterion. The error bounds presented improve upon the previously available ones in the aspects of both providing a more accurate estimate of the identification error as well as establishing a faster convergence rate for the proposed algorithm.     

Autoregressive and adaptive estimation with an application to hurricane track prediction

The key result of this paper is that following a change in a parameter of AR (p), an autoregressive process of order p, the innovations sequence of the Kalman filter parameters will follow an autoregressive moving-average model in addition to a transient function. Furthermore, it is also shown that the first p values of the innovations' sample autocorrelation can be used to form a sufficient statistic to detect if at least one of the parameters in the AR (p) model did change at an unknown point in time. Following a parameter change detection process, improved estimates and noise statistics can be determined and implemented to modify the Kalman filter. The revised model will thus be more consistent with the most recent process behaviour. To motivate the reader, a simulation exercise was conducted to validate the on-line change detector and adaptive estimation algorithm. The proposed algorithm was used to predict hurricane movements with real data provided by the National Hurricane Center.     

A Posteriori Error Estimates and an Adaptive Finite Element Method for the Allen–Cahn Equation and the Mean Curvature Flow

This paper develops an a posteriori error estimate of residual type for finite element approximations of the Allen–Cahn equation u_t − Δu+ ε⁻² f(u)=0. It is shown that the error depends on ε⁻¹ only in some low polynomial order, instead of exponential order. Based on the proposed a posteriori error estimator, we construct an adaptive algorithm for computing the Allen–Cahn equation and its sharp interface limit, the mean curvature flow. Numerical experiments are also presented to show the robustness and effectiveness of the proposed error estimator and the adaptive algorithm.     

Time domain identification for robust control

In this paper, we are concerned with a problem of robust control-oriented system identification in the time domain. Based on the well-known Schur-Takagi-AAK Theorem, we propose a linear algorithm to obtain the nominal model of the plant to be identified and the minimal bound of the uncertainty of the nominal model error which is measured by H^∞-norm. It is also shown that, in the model set defined by the nominal model and the uncertainty bound, there exists at least one model which matches the prescribed input-output data given in the time domain.     

Toward a theory of Pollard's rho method

Pollard's “rho” method for integer factorization iterates a simple polynomial map and produces a nontrivial divisor of n when two such iterates agree modulo this divisor. Experience and heuristic arguments suggest that a prime divisor p should be detected in steps, but this has never been proved. Indeed, nothing seems to be have been rigorously proved about the probability of success that would improve the obvious lower bound of 1/p. This paper shows that for fixed k, this probability is at least (₂^k)/p + O(p^−3/2) as p → ∞. This leads to an Ω(log²p)/p estimate of the success probability.     

Fast Parallel Algorithms for Solving Triangular Systems of Linear Equations on the Hypercube

This paper presents efficient hypercube algorithms for solving triangular systems of linear equations by using various matrix partitioning and mapping schemes. Recently, several parallel algorithms have been developed for this problem. In these algorithms, the triangular solver is treated as the second stage of Gauss elimination. Thus, the triangular matrix is distributed by columns (or rows) in a wrap fashion since it is likely that the matrix is distributed this way after an LU decomposition has been done on the matrix. However, the efficiency of the algorithms is low. Our motivation is to develop various data partitioning and mapping schemes for hypercube algorithms by treating the triangular solver as an independent problem. Theoretically, the computation time of our best algorithm is ((12p + 1)n² + 36p³ − 28p²)/(24p²), and an upper bound on the communication time is 2αp log p (log n − log p) + 2α(log n − log p − 1) log p + (cn/p − 2c)(2 log p − 1) + log p(cn − c − α), where α is the (communication startup time)/(one entry scanning time), c is a constant, n is the order of the triangular system and p is the number of nodes in the hypercube. Experimental results show that the algorithm is efficient. The efficiency of the algorithm is 0.945 when p = 2, n = 513, and 0.93 when p = 8, n = 1025.     

Performing work with asynchronous processors: Message-delay-sensitive bounds

This paper considers the problem of performing tasks in asynchronous distributed settings. This problem, called Do-All, has been substantially studied in synchronous models, but there is a dearth of efficient algorithms for asynchronous message-passing processors. Do-All can be trivially solved without any communication by an algorithm where each processor performs all tasks. Assuming p processors and t tasks, this requires work Θ (p · t). Thus, it is important to develop subquadratic solutions (when p and t are comparable) by trading computation for communication. Following the observation that it is not possible to obtain subquadratic work when the message delay d is substantial, e.g., d = Θ (t), this work pursues a message-delay-sensitive approach. Here, the upper bounds on work and communication are given as functions of p, t, and d, the upper bound on message delays, however, algorithms have no knowledge of d and they cannot rely on the existence of an upper bound on d. This paper presents two families of asynchronous algorithms achieving, for the first time, subquadratic work as long as d = o (t). The first family uses as its basis a shared-memory algorithm without having to emulate atomic registers assumed by that algorithm. These deterministic algorithms have work O (tp^ε + pdt/d^ε) for any ε > 0. The second family uses specific permutations of tasks, with certain combinatorial properties, to sequence the work of the processors. These randomized (deterministic) algorithms have expected (worst-case) work O (t log p + pd log (2 + t/d)). Another important contribution in this work is the first delay-sensitive lower bound for this problem that helps explain the behavior of our algorithms: any randomized (deterministic) algorithm has expected (worst-case) work of Ω (t + pd log_d+1t).     

Exact filters for Newton–Raphson parameter estimation algorithms for continuous-time partially observed stochastic systems

This paper presents explicit finite-dimensional filters for implementing Newton–Raphson (NR) parameter estimation algorithms. The models which exhibit nonlinear parameter dependence are stochastic, continuous-time and partially observed. The implementation of the NR algorithm requires evaluation of the log-likelihood gradient and the Fisher information matrix. Fisher information matrices are important in bounding the estimation error from below, via the Cramer–Rao bound. The derivations are based on relations between incomplete and complete data, likelihood, gradient and Hessian likelihood functions, which are derived using Girsanov's measure transformations.     

基于神经网络的一类生化反应器状态在线估计

提出一种随机非线性系统状态和参数同时估计的神经网络新方法，并证明了该方法的无偏性和是小方差性，将其用于乙醇间歇发酵器的状态和参数估计，结果表明估计值民实验值相吻合，此方法对噪声特片无特殊要求，对初始状态估值不敏感，对初始参数值具有一定的鲁棒性，可利用有限的状态量测信息在线估计不可测量的状态变量和物理参数。