A hybrid algorithm to optimize RBF network architecture and parameters for nonlinear time series prediction

This paper proposes a novel hybrid algorithm for automatic selection of the proper input variables, the number of hidden nodes of the radial basis function (RBF) network, and optimizing network parameters (weights, centers and widths) simultaneously. In the proposed algorithm, the inputs and the number of hidden nodes of the RBF network are represented by binary-coded strings and evolved by a genetic algorithm (GA). Simultaneously, for each chromosome with fixed inputs and number of hidden nodes, the corresponding parameters of the network are real-coded and optimized by a gradient-based fast-converging parameter estimation method. Performance of the presented hybrid approach is evaluated by several benchmark time series modeling and prediction problems. Experimental results show that the proposed approach produces parsimonious RBF networks, and obtains better modeling accuracy than some other algorithms.     

ARMAX时间序列模型异常点及异常点斑片的估计和检测

将通常的Gibbs抽样和自适应的Gibbs抽样算法用于带有外生变量的自回归移动平均时间序列(ARMAX)模型的Bayes分析,首先采用一些方法消除ARMAX模型中输入(外生变量)序列的影响,然后在前人工作的基础上给出了一种类似的挖掘相应时间序列中的异常点及异常点斑片的方法.说明了自适应的Gibbs抽样算法也能够有效地检测ARMAX模型中孤立的附加型异常点及异常点斑片.实际的和模拟的结果也显示这些方法可以明显减少掩盖和淹没现象的发生,这是对已有工作的推广和扩充.     

Frequency domain characteristics of linear operator to decompose a time series into the multi-components

Frequency domain properties of the operators to decompose a time series into the multi-components along the Akaike's Bayesian model (Akaike (1980, Bayesian Statistics, 143–165, University Press, Valencia, Spain)) are shown. In that analysis a normal disturbance-linear-stochastic regression prior model is applied to the time series. A prior distribution, characterized by a small number of hyperparameters, is specified for model parameters. The posterior distribution is a linear function (filter) of observations. Here we use frequency domain analysis or filter characteristics of several prior models parametrically as a function of the hyperparameters.     

Forecasting time series with multiple seasonal patterns

A new approach is proposed for forecasting a time series with multiple seasonal patterns. A state space model is developed for the series using the innovations approach which enables us to develop explicit models for both additive and multiplicative seasonality. Parameter estimates may be obtained using methods from exponential smoothing. The proposed model is used to examine hourly and daily patterns in hourly data for both utility loads and traffic flows. Our formulation provides a model for several existing seasonal methods and also provides new options, which result in superior forecasting performance over a range of prediction horizons. In particular, seasonal components can be updated more frequently than once during a seasonal cycle. The approach is likely to be useful in a wide range of applications involving both high and low frequency data, and it handles missing values in a straightforward manner.     

The effects of advertising,prices and distribution on market share volatility

Although the influence of marketing decision variables on market share has received ample attention in the literature, less is known about their effects on volatility. This study attempts to shed light on this issue by empirically examining the effects of advertising, pricing and distribution, not only on market share but also on its volatility, using an EGARCH model. We argue that establishing a link between management-controlled actions, such as advertising, pricing and distribution, and market share volatility may benefit firms seeking to reduce uncertainty in their market share performance. Application of the proposed model to data on two markets (SUVs and Minivans), suggested that advertising, pricing and distribution significantly influence market share volatility.     

Visualizing the Relationship between Two Time Series by Hierarchical Smoothing Models

Abstract

Let x(t_i), y(t_i) be two time series such that y(t_i) = μ(t_i, x) + ε_i, where μ is a smooth function and ε_i is a zero mean stationary process. Which model may be assumed for μ depends on the subject specific context. This article was motivated by questions raised in the context of musical performance theory. The general problem is to understand the relationship between the symbolic structure of a music score and its performance. Musical structure typically consists of a hierarchy of global and local structures. This motivates the definition of hierarchical smoothing models (or HISMOOTH models) that are characterized by a hierarchy of bandwidths b ₁ > b ₂ > … > b_M and a vector of coefficients β ∈ R^M. The expected value μ(t_i x) = E[y(t_i)‖x] is equal to a weighted sum of smoothed versions of x. The “errors” ε_i are modeled by a Gaussian process that may exhibit long memory. More generally, we may observe a collection of time series y_r (r = 1, …, N) that are related to a common time series x by y_r(t_i) = μ _r(t_i, x) + ε_{r, i} where ε _r are independent error processes. For repeated time series, HISMOOTH models lead to a visual and formal classification into clusters that can be interpreted in terms of the relationship to x. An analysis of tempo curves from 28 performances of Schumann's “Träumerei” op. 15/7 illustrates the method. In particular, similarities and differences of “melodic styles” can be identified.     

稳健残差控制图的构建及在金融时序中的应用

对于呈现自相关和波动族聚性并存的受控过程,通常采用残差控制图对其进行监控。但异常点的存在会对自相关或波动族聚性模型的拟合产生重要影响,使得基于该模型的残差并非独立同分布导致常规残差控制图监控失效。为解决这类问题,本文提出稳健残差控制图。即建立稳健的ARMA模型解决自相关问题从而得到无自相关的残差序列,用稳健的GARCH模型来构建控制图的上下限。模拟和实证研究表明,本文提出的稳健残差控制图具有很好的抗异常点能力并能更好的对金融时间序列的异常现象进行监控。     

Riccati方程与(2+1)维非线性物理模型的分离变量解

对于(2+1)维非线性物理模型,欲获其解并非易事.本文试图在变量分离法中,通过设定的R iccati方程来得到方程的解.同时,以(2+1)维Bo iti-L eon-M anna-P em p inelli方程和Burgers方程为例来说明之.     

A new time-discretization for delay multiple-input nonlinear systems using the Taylor method and first order hold

A new discretization method is proposed for multi-input-driven nonlinear continuous systems with time-delays, based on a combination of the Taylor series expansion and the first-order hold (FOH) assumption. The mathematical structure of the new discretization scheme is explored. On the basis of this structure, the sampled-data representation of the time-delayed multi-input nonlinear system is derived. First the new approach is applied to nonlinear systems with two inputs, and then the delayed multi-input general equation is derived. The resulting time discretization method provides a finite-dimensional representation for multi-input nonlinear systems with time-delays, thereby enabling the application of existing controller design techniques to such systems. The performance of the proposed method is evaluated using a nonlinear system with time-delays (maneuvering an automobile). Various sampling rates, time-delay values and control inputs are considered to evaluate the proposed method. The results demonstrate that the proposed discretization scheme can meet the system requirements even when using a large sampling period with precision limitations. The discretization results of the FOH method are also compared with those of the zero order hold (ZOH) method. The precision of the FOH method in the discretization procedure combined with the Taylor series expansion is much higher than that of the ZOH method except in the case of constant inputs.     

Nonlinear Filter Estimation of Volatility

A discrete time nonlinear filter is used to estimate the volatility in a financial model. New filters are derived for sums of unobserved quantities and the EM algorithm applied to determine the parameters of the model.     

厚尾金融时间序列的贝叶斯单位根检验

在金融时间序列分析中,单位根检验是一个相当重要的研究问题。对于数据生成过程为自回归的厚尾金融时序数据,古典的ADF单位根检验统计量应用非常困难。本文则在贝叶斯框架下,发展了检验带有未知自由度厚尾t分布的自回归金融时间序列单位根的贝叶斯方法。蒙特卡罗模拟结果显示本文发展的方法能够取得好的检验功效。最后,用万科地产月度历史收益数据来演示了本文发展的方法。     

A Retrospective Study on Applications of the Lindley Distribution

The need for efficient statistical models has increased with the flow of new data, which makes distribution theory a particularly interesting and attractive field. Here, we provide a thorough study of the applications of the Lindley distribution and its diverse generalizations. More precisely, we review some special applications in various areas, such as time series analysis, stress strength analysis, acceptance sampling plans and data analysis. We also conduct a comparative study between the Lindley distribution and some of its generalizations by using four real-life data sets.     

Computational aspects of robust Holt-Winters smoothing based on M-estimation

To obtain a robust version of exponential and Holt-Winters smoothing the idea of M-estimation can be used. The difficulty is the formulation of an easy-to-use recursive formula for its computation. A first attempt was made by Cipra (Robust exponential smoothing, J. Forecast. 11 (1992), 57–69). The recursive formulation presented there, however, is unstable. In this paper, a new recursive computing scheme is proposed. A simulation study illustrates that the new recursions result in smaller forecast errors on average. The forecast performance is further improved upon by using auxiliary robust starting values and robust scale estimates. This research has been supported by the Research Fund K.U. Leuven and the Fonds voor Wetenschappelijk Onderzoek (Contract number G.0594.05).     

Complexity Penalized M-Estimation

We present very fast algorithms for the exact computation of estimators for time series, based on complexity penalized log-likelihood or M-functions. The algorithms apply to a wide range of functionals with morphological constraints, in particular to Potts or Blake–Zisserman functionals. The latter are the discrete versions of the celebrated Mumford–Shah functionals. All such functionals contain model parameters. Our algorithms allow for optimization not only for each separate parameter, but even for all parameters simultaneously. This allows for the examination of the models in the sense of a family approach. The algorithms are accompanied by a series of illustrative examples from molecular biology.     

Bayesian approach to global optimization and application to multiobjective and constrained problems

In this paper, the Bayesian methods of global optimization are considered. They provide the minimal expected deviation from the global minimum. It is shown that, using the Bayesian methods, the asymptotic density of calculations of the objective function is much greater around the point of global minimum. The relation of this density to the parameters of the method and to the function is defined.Algorithms are described which apply the Bayesian methods to problems with linear and nonlinear constraints. The Bayesian approach to global multiobjective optimization is defined. Interactive procedures and reduction of multidimensional data in the case of global optimization are discussed.     

A Class of Sparse Invertible Matrices and Their Use for Nonlinear Prediction of Nearly Periodic Time Series with Fixed Period

We introduce a class of sparse matrices U _m (A _{p 1} ) of order m by m, where m is a composite natural number, p ₁ is a divisor of m, and A _{p 1} is a set of nonzero real numbers of length p ₁. The construction of U _m (A _{p 1} ) is achieved by iteration, involving repetitive dilation operations and block-matrix operations. We prove that the matrices U _m (A _{p 1} ) are invertible and we compute the inverse matrix (U _m (A _{p 1} ))^?1 explicitly. We prove that each row of the inverse matrix (U _m (A _{p 1} ))^?1 has only two nonzero entries with alternative signs, located at specific positions, related to the divisors of m. We use the structural properties of the matrix (U _m (A _{p 1} ))^?1 in order to build a nonlinear estimator for prediction of nearly periodic time series of length m with fixed period.     

Analysis of event-based, single-server nonstationary simulation responses using classical time-series models

In this article, we present a metamodeling methodology for analyzing event-based, single-server nonstationary simulation responses that is based on the use of classical ARIMA (or SARIMA) time-series models. Some analytical results are derived for a Markovian queue and are used to evaluate the proposed methodology. The use of the corresponding procedure is illustrated on a traffic example from the simulation literature. Some conclusions are drawn and recommendations for further work are stated.     

Measurement of multiperiod aggregative efficiency

This article proposes a new method for measuring an aggregative efficiency of multiple period production systems. Every organization or firm generates a time series of data that represent its performances in the resource utilization and output production over multiple periods, and often desires an aggregated measure of efficiency for several periods of interest. Data envelopment analysis (DEA) has become an accepted and well-known approach to evaluating efficiency performance in a wide range of cases. However, most of the DEA studies have dealt primarily with ways to gauge the efficiency of production in only a single period so this efficiency reflects the insufficient or partial performance of multiple period productions. The new method is developed through extensions of the concept of Debreu–Farrell technical efficiency and is applied to evaluating the efficiency of cable TV service units with 3-year data.     

Aggregation of spectral density estimators

Given stationary time series data, we study the problem of finding the best linear combination of a set of lag window spectral density estimators with respect to the mean squared risk. We present an aggregation procedure and prove a sharp oracle inequality for its risk. We also provide simulations demonstrating the performance of our aggregation procedure, given Bartlett and other estimators of varying bandwidths as input. This extends work by P. Rigollet and A. Tsybakov on aggregation of density estimators.