A new algorithm for linear and nonlinear ARMA model parameter estimation using affine geometry

A linear and nonlinear autoregressive (AR) moving average (MA) (ARMA) identification algorithm is developed for modeling time series data. The new algorithm is based on the concepts of affine geometry in which the salient feature of the algorithm is to remove the linearly dependent ARMA vectors from the pool of candidate ARMA vectors. For noiseless time series data with a priori incorrect model-order selection, computer simulations show that accurate linear and nonlinear ARMA model parameters can be obtained with the new algorithm. Many algorithms, including the fast orthogonal search (FOS) algorithm, are not able to obtain correct parameter estimates in every case, even with noiseless time series data, because their model-order search criteria are suboptimal. For data contaminated with noise, computer simulations show that the new algorithm performs better than the FOS algorithm for MA processes, and similarly to the FOS algorithm for ARMA processes. However, the computational time to obtain the parameter estimates with the new algorithm is faster than with FOS. Application of the new algorithm to experimentally obtained renal blood flow and pressure data show that the new algorithm is reliable in obtaining physiologically understandable transfer function relations between blood pressure and flow signals.     

Profiled corrugated circular horns analysis and synthesis via anartificial neural network

In many reflector and lens antennas profiled corrugated circular horns constitute one of the best feed solution thanks to their polarization purity and small size. In this paper, a method for the design of these feeds by using an artificial neural network (ANN) approach is described. The results obtained with such an approach are investigated both for the analysis and for the synthesis problem, and compared with the standard methods. This unconventional solution gives a good level of accuracy and shorter processing times especially for the synthesis problem, where there is still a lack of affordable and fully automated procedures     

非线性参数估计中的观测集预处理技术

非线性参数估计模型中,若系统的可观测度较弱,被估计参数相互耦合,那么参数的估计精度不仅与随机观测噪声有关,与系统的可观测度也有着密切的关系。该文提出一种观测集预处理方法,依据可观测度指标对观测数据进行筛选,然后利用筛选集对参数进行估计。以雷达的系统误差估计为例,使用筛选集估计雷达的系统误差,比直接使用原观测集估计所得结果更为精确。     

An analog architecture on parameter estimation of ARMA models

An analog architecture that is suitable for parameter estimation of autoregressive moving average (ARMA) models is proposed. The convergence theorem that connects this architecture with ARMA parameter estimation is presented. Simulation results indicate that its convergence takes only a few microseconds. Hence, this architecture can lead to online implementations     

基于小波神经网络和ARMA模型的CPI波动区间预测

目前对于消费者价格指数(CPI)的预测研究基本集中于点预测。为预测本期较上期的CPI数据波动区间,提出一种基于小波神经网络和ARMA组合模型预测的方法。该模型首先利用小波神经网络对CPI数据进行拟合测试,对测试序列实际输出和期望输出的残差序列{et}进行ARMA建模预测,然后基于方差最小原则得到预测残差序列{e!t}95%的置信区间。通过实验表明预测残差序列95%置信区间可以很好的反应未来CPI数据的波动情况,具有较高的参考价值。     

Approximating nonlinear fading-memory operators using neural network models

A fading-memory system is defined as a system whose response map has unique asymptotic properties over some set of inputs. It is shown that any discrete-time fading-memory system can be uniformly approximated arbitrarily closely over a compact set of input sequences by uniformly approximating either its external or internal representation sufficiently closely. In other words, the problem of uniformly approximating a fading-memory system reduces to the problem of uniformly approximating continuous real-valued functions on compact sets. The perceptron is shown to realize a set of continuous real-valued functions that is uniformly dense on compacta in the set of all continuous functions. Using the perceptron to uniformly approximate the external and internal representations of a fading-memory system results, respectively, in simple nonlinear finite-memory and infinite-memory system models.     

Joint state and parameter estimation for a target-directed nonlinear dynamic system model

We present a new approach to joint state and parameter estimation for a target-directed, nonlinear dynamic system model with switching states. The model, recently proposed for representing speech dynamics, is called the hidden dynamic model (HDM). The model parameters, subject to statistical estimation, consist of the target vector and the system matrix (also called "time-constants"), as well as parameters characterizing the nonlinear mapping from the hidden state to the observation. We implement these parameters as the weights of a three-layer feedforward multilayer perceptron (MLP) network. The new estimation approach is based on the extended Kalman filter (EKF), and its performance is compared with the traditional expectation-maximization (EM) based approach. Extensive simulation results are presented using both approaches and under typical HDM speech modeling conditions. The EKF-based algorithm demonstrates superior convergence performance compared with the EM algorithm, but the former suffers from excessive computational loads when adopted for training the MLP weights. In all cases, the simulated model output converges to the given observation sequence. However, only in the case where the MLP weights or the target vector are assumed known do the time-constant parameters converge to their true values. We also show that the MLP weights never converge to their true values, thus demonstrating the many-to-one mapping property of the feedforward MLP. We conclude that, for the system to be identifiable, restrictions on the parameter space are needed.     

Using neural network based nonlinear IFS to model time sequences

A novel neural network based iterated function system (IFS) model is presented in this paper while the precondition to ensure the model is also explored. Applying it to some practical data, the given signal can be approximated exactly by the attractor generated by this model, which provides another way to resolve fractal inverse problem.     

Nonlinear autoregressive and nonlinear autoregressive moving average model parameter estimation by minimizing hypersurface distance

The least squares (LS) can be used for nonlinear autoregressive (NAR) and nonlinear autoregressive moving average (NARMA) parameter estimation. However, for nonlinear cases, the LS results in biased parameter estimation due to its assumption that the independent variables are noise free. The total least squares (TLS) is another method that can used for nonlinear parameter estimation to increase the accuracy of the LS because it specifically accounts for the fact that the independent variables are noise corrupted. TLS has its own limitations, however, mainly because it is difficult for the method to isolate noise from the signal components. We present a new method that is based on minimization of hypersurface distance for accurate parameter estimation for NAR and NARMA models. Computer simulation examples show that the new method results in far more accurate NAR and NARMA model parameter estimates than do either the LS and TLS, with noise that is either white or colored, and retains its high accuracy even when the signal-to-noise ratio (SNR) is as low as 10 dB.     

Maximum likelihood parameter estimation of textures using aWold-decomposition based model

We present a solution to the problem of modeling, parameter estimation, and synthesis of natural textures. The texture field is assumed to be a realization of a regular homogeneous random field, which can have a mixed spectral distribution. On the basis of a 2-D Wold-like decomposition, the field is represented as a sum of a purely indeterministic component, a harmonic component, and a countable number of evanescent fields. We present a maximum-likelihood solution to the joint parameter estimation problem of these components from a single observed realization of the texture field. The proposed solution is a two-stage algorithm. In the first stage, we obtain an estimate for the number of harmonic and evanescent components in the field, and a suboptimal initial estimate for the parameters of their spectral supports. In the second stage, we refine these initial estimates by iterative maximization of the likelihood function of the observed data. By introducing appropriate parameter transformations the highly nonlinear least-squares problem that results from the maximization of the likelihood function, is transformed into a separable least-squares problem. The solution for the unknown spectral supports of the harmonic and evanescent components reduces the problem of solving for the transformed parameters of the field to a linear least squares. Solution of the transformation equations then provides a complete solution of the field-model parameter estimation problem. The Wold-based model and the resulting analysis and synthesis algorithms are applicable to a wide variety of texture types found in natural images.     

Building neural network equipment models using model modifiertechniques

In this paper, we address the problem of developing accurate neural network equipment models economically. To this end, we propose model modifier techniques in conjunction with physical-neural network models. Two model modifiers-difference method and source input method-are proposed and evaluated on a horizontal chemical vapor deposition reactor. The results show that the source input method outperforms the difference method. Further, to develop a model of comparable accuracy, the source input method reduces the number of experimental data points to approximately one fourth of those needed without this approach     

Optimal ARMA parameter estimation based on the sample covariancesfor data with missing observations

The problem of spectral estimation through the autoregressive moving-average (ARMA) modeling of stationary processes with missing observations is considered. A class of estimators based on the sample covariances is presented, and an asymptotically optimal estimator in this class is proposed. The proposed algorithm is based on a nonlinear-least-squares fit of the sample covariances computed from the data to the true covariances of the assumed ARMA model. The statistical properties of the algorithm are explored and used to show that it is asymptotically optimal, in the sense of achieving the smallest possible asymptotic variance. The performance of the algorithm is illustrated by some numerical examples     

Time-varying ARMA stable process estimation using sequential Monte Carlo

Various time series data in applications ranging from telecommunications to financial analysis and from geophysical signals to biological signals exhibit non-stationary and non-Gaussian characteristics. α-Stable distributions have been popular models for data with impulsive and non-symmetric characteristics. In this work, we present time-varying autoregressive moving-average α-stable processes as a potential model for a wide range of data, and we propose a method for tracking the time-varying parameters of the process with α-stable distribution. The technique is based on sequential Monte Carlo, which has assumed a wide popularity in various applications where the data or the system is non-stationary and non-Gaussian.     

Robust parameter estimation for mixture model

In pattern recognition, when the ratio of the number of training samples to the dimensionality is small, parameter estimates become highly variable, causing the deterioration of classification performance. This problem has become more prevalent in remote sensing with the emergence of a new generation of sensors with as many as several hundred spectral bands. While the new sensor technology provides higher spectral and spatial resolution, enabling a greater number of spectrally separable classes to be identified, the needed labeled samples for designing the classifier remain difficult and expensive to acquire. Better parameter estimates can be obtained by exploiting a large number of unlabeled samples in addition to training samples, using the expectation maximization algorithm under the mixture model. However, the estimation method is sensitive to the presence of statistical outliers. In remote sensing data, miscellaneous classes with few samples are often difficult to identify and may constitute statistical outliers. Therefore, the authors propose to use a robust parameter-estimation method for the mixture model. The proposed method assigns full weight to training samples, but automatically gives reduced weight to unlabeled samples. Experimental results show that the robust method prevents performance deterioration due to statistical outliers in the data as compared to the estimates obtained from the direct EM approach     

Deep face liveness detection based on nonlinear diffusion using convolution neural network

A face-spoofing attack occurs when an imposter manipulates a face recognition and verification system to gain access as a legitimate user by presenting a 2D printed image or recorded video to the face sensor. This paper presents an efficient and non-intrusive method to counter face-spoofing attacks that uses a single image to detect spoofing attacks. We apply a nonlinear diffusion based on an additive operator splitting scheme. Additionally, we propose a specialized deep convolution neural network that can extract the discriminative and high-level features of the input diffused image to differentiate between a fake face and a real face. Our proposed method is both efficient and convenient compared with the previously implemented state-of-the-art methods described in the literature review. We achieved the highest reported accuracy of 99% on the widely used NUAA dataset. In addition, we tested our method on the Replay Attack dataset which consists of 1200 short videos of both real access and spoofing attacks. An extensive experimental analysis was conducted that demonstrated better results when compared to previous static algorithms results. However, this result can be improved by applying a sparse autoencoder learning algorithm to obtain a more distinguishable diffused image.     

Linear parameter estimation for induction machines considering theoperating conditions

The use of linear parameter estimation techniques to determine the stator resistance, self-inductance of the stator winding, transient inductance, rotor time constant, as well as the angular shaft speed of a three-phase induction machine is investigated in this paper. In order to obtain results with maximum accuracy, some specific procedures to reduce the effect of the operating conditions on the quality of the estimates are investigated. Both computer and experimental results are used to anchor the main conclusions issued from this study     

Thermal parameter estimation using recursive identification

A novel method that converts a semiconductor transient thermal impedance curve (TTIC) into an equivalent thermal RC network model is presented. Thermal resistance (R) and thermal capacitance (C) parameters of the model are identified using manufacturer's data and offline recursive least square techniques. Relevant estimation theory concepts and the formulation of an appropriate model for the identification process are given. Model synthesis is illustrated using an isolated base power transistor module. The application of time decoupled theory for high order thermal models is outlined. Simulation of junction temperature responses using model and manufacturer TTICs are compared. Estimated parameter validity is further confirmed by parameter calculation obtained from module physical dimensions     

Research onconvolutional neural network for reservoir parameter prediction

As the branch of artificial intelligence,artificial neural network solved many difficult practical problems in pattern recognition and classification prediction field successfully.However,they cannot learn the feature from networks.In recent years,deep learning becomes more and more advanced,but the research on the field of geological reservoir pa-rameter prediction is still rare.A method to predict reservoir parameters by convolutional neural network was presented,which can not only predict reservoir parameters accurately,but also get features of the geological reservoir.The study es-tablished the convolutional neural network model.Results show that the convolutional neural network can be used for reservoir parameter prediction,and get high prediction precision.Moreover,convolutional features from convolutional neural network provided important support for geological modeling and logging interpretation.     

Hierarchical neural network for hand pose estimation

Hand pose estimation plays an important role in human–computer interaction and augmented reality. Regressing the joints coordinates is a difficult task due to the flexibility of the joint, self-occlusion and so on. In this paper, we propose a novel and simple hierarchical neural network for hand pose estimation. The hand joint coordinates are divided into six parts and each part is regressed in sequence with this hierarchical architecture. This can divide the complex task of regressing all hand joints coordinates into several sub-tasks which can make the estimation more accurate. When regress the joint coordinates of one part, the features of other parts may bring negative influence to this part due to the similarity among the fingers, so we use an interference cancellation operation in our hierarchical architecture. At the time the joint coordinates of one part are regressed, the corresponding features will be removed from the hand global feature to eliminate the interference of this part. The obtained features will be used as input for regressing the joints coordinates of the next part. The ablation study verifies the effectiveness of our hierarchical architecture. The experimental results demonstrate that our method can achieve state-of-the-art or comparable results relative to existing methods on four public hand pose datasets.     

Robust time delay estimation of bioelectric signals using least absolute deviation neural network

The time delay estimation (TDE) is an important issue in modern signal processing and it has found extensive applications in the spatial propagation feature extraction of biomedical signals as well. Due to the extreme complexity and variability of the underlying systems, biomedical signals are usually nonstationary, unstable and even chaotic. Furthermore, due to the limitations of the measurement environments, biomedical signals are often noise-contaminated. Therefore, the TDE of biomedical signals is a challenging issue. A new TDE algorithm based on the least absolute deviation neural network (LADNN) and its application experiments are presented in this paper. The LADNN is the neural implementation of the least absolute deviation (LAD) optimization model, also called unconstrained minimum L1-norm model, with a theoretically proven global convergence. In the proposed LADNN-based TDE algorithm, a given signal is modeled using the moving average (MA) model. The MA parameters are estimated by using the LADNN and the time delay corresponds to the time index at which the MA coefficients have a peak. Due to the excellent features of L1-norm model superior to Lp-norm (p > 1) models in non-Gaussian noise environments or even in chaos, especially for signals that contain sharp transitions (such as biomedical signals with spiky series or motion artifacts) or chaotic dynamic processes, the LADNN-based TDE is more robust than the existing TDE algorithms based on wavelet-domain correlation and those based on higher-order spectra (HOS). Unlike these conventional methods, especially the current state-of-the-art HOS-based TDE, the LADNN-based method is free of the assumption that the signal is non-Gaussian and the noises are Gaussian and, thus, it is more applicable in real situations. Simulation experiments under three different noise environments, Gaussian, non-Gaussian and chaotic, are conducted to compare the proposed TDE method with the existing HOS-based method. Real application experiment is conducted to extract time delay information between every two adjacent channels of gastric myoelectrical activity (GMA) to assess the spatial propagation characteristics of GMA during different phases of the migrating myoelectrical complex (MMC).