A neural networks-based negative selection algorithm in fault diagnosis

Inspired by the self/nonself discrimination theory of the natural immune system, the negative selection algorithm (NSA) is an emerging computational intelligence method. Generally, detectors in the original NSA are first generated in a random manner. However, those detectors matching the self samples are eliminated thereafter. The remaining detectors can therefore be employed to detect any anomaly. Unfortunately, conventional NSA detectors are not adaptive for dealing with time-varying circumstances. In the present paper, a novel neural networks-based NSA is proposed. The principle and structure of this NSA are discussed, and its training algorithm is derived. Taking advantage of efficient neural networks training, it has the distinguishing capability of adaptation, which is well suited for handling dynamical problems. A fault diagnosis scheme using the new NSA is also introduced. Two illustrative simulation examples of anomaly detection in chaotic time series and inner raceway fault diagnosis of motor bearings demonstrate the efficiency of the proposed neural networks-based NSA.     

On input-output linearization of discrete-time nonlinear systems

We consider a nonlinear discrete-time system of the form Σ: x(t+1)=f(x(t), u(t)), y(t) =h(x(t)), where x ε R^N, u ε R^m, y ε R^q and f and h are analytic. Necessary and sufficient conditions for local input-output linearizability are given. We show that these conditions are also sufficient for a formal solution to the global input-output linearization problem. Finally, we show that zeros at infinity of ε can be obtained by the structure algorithm for locally input-output linearizable systems.     

Improving neural network models of physical systems through dimensional analysis

The paper presents a technique for generating concise neural network models of physical systems. The neural network models are generated through a two-stage process. The first stage uses information embedded in the dimensions or units in which the data is represented. Dimensional analysis techniques are used initially to make this information explicit, and a limited search in the neural network architecture space is then conducted to determine dimensionless representations of variables/parameters that perform well for a given model complexity. The second stage uses information available in the numerical values of the data to search for high-level dimensionless variables/parameters, generated from simple combinations of dimensionless quantities generated in the first stage and which result in concise neural network models with improved performance characteristics. The search for these high-level dimensionless variables/parameters is conducted in an enhanced representation space using functional link networks with flat or near flat architectures. The use and effectiveness of the technique is demonstrated for three applications. The first is the design and analysis of reinforced concrete beams, which is representative of the class of problems associated with the design and analysis of composites. The second is the classical elastica problem, for predicting non-linear post-buckled behaviour of columns and the third, the analysis of a bent bar under a specified combination of loads.     

Discrete-time neural network output feedback control of nonlinear discrete-time systems in non-strict form

An adaptive neural network (NN)-based output feedback controller is proposed to deliver a desired tracking performance for a class of discrete-time nonlinear systems, which are represented in non-strict feedback form. The NN backstepping approach is utilized to design the adaptive output feedback controller consisting of: (1) an NN observer to estimate the system states and (2) two NNs to generate the virtual and actual control inputs, respectively. The non-causal problem encountered during the control design is overcome by using a dynamic NN which is constructed through a feedforward NN with a novel weight tuning law. The separation principle is relaxed, persistency of excitation condition (PE) is not needed and certainty equivalence principle is not used. The uniformly ultimate boundedness (UUB) of the closed-loop tracking error, the state estimation errors and the NN weight estimates is demonstrated. Though the proposed work is applicable for second order nonlinear discrete-time systems expressed in non-strict feedback form, the proposed controller design can be easily extendable to an nth order nonlinear discrete-time system.     

生物视觉系统的神经网络编码模型综述

生物视觉系统的研究一直是计算机视觉算法的重要灵感来源。有许多计算机视觉算法与生物视觉研究具有不同程度的对应关系,包括从纯粹的功能启发到用于解释生物观察的物理模型的方法。从视觉神经科学向计算机视觉界传达的经典观点是视觉皮层分层层次处理的结构。而人工神经网络设计的灵感来源正是视觉系统中的分层结构设计。深度神经网络在计算机视觉和机器学习等领域都占据主导地位。许多神经科学领域的学者也开始将深度神经网络应用在生物视觉系统的计算建模中。深度神经网络多层的结构设计加上误差的反向传播训练,使得它可以拟合绝大多数函数。因此,深度神经网络在学习视觉刺激与神经元响应的映射关系并取得目前性能最好的模型同时,网络内部的单元甚至学习出生物视觉系统子单元的表达。本文将从视网膜等初级视觉皮层和高级视觉皮层(如,视觉皮层第4区(visual area 4,V4)和下颞叶皮层(inferior temporal,IT))分别介绍基于神经网络的视觉系统编码模型。主要内容包括:1)有关视觉系统模型的概念与定义;2)初级视觉系统的神经网络预测模型;3)任务驱动的高级视觉皮层编码模型。最后本文还将介绍最新有关无监督学习的神经编码...     

Multiple models and neural networks based decoupling control of ball mill coal-pulverizing systems

Using a ball mill coal-pulverizing system as a motivating/application example, a class of complex industrial processes is investigated in this paper, which has strong couplings among loops, high nonlinearities and time-varying dynamics under different operation conditions. Focusing on such processes, an intelligent decoupling control method is developed, where the effects of nonlinearities are dealt with by neural network compensations and coupling effects are handled by specifically designed decoupling compensators, while the effect of time-varying dynamics is treated by a switching mechanism among multiple models. The stability and convergence of the closed-loop system are analyzed. The proposed method has been applied to the ball mill coal-pulverizing systems of 200 MW units in a heat power plant in China. Application results show that the system outputs are maintained in desired scopes, the electric energy consumption per unit coal has been reduced by 10.3%, and the production rate has been increased by 8%.     

A comparative study for forecasting using neural networks vs genetically identified Box&Jenkins models

This paper aims to discuss the results and conclusions of an extensive comparative study on the forecasting performance between two different techniques: a genetic expert system in which a genetic algorithm carries out the identification stage embraced in the three- phase Box&Jenkins univariate methodology; and a connectionist approach. At the heart of the former, an expert system rules the identification-estimation-diagnostic checking cyclical process to end up with the predictions provided by the SARIMA model which best fits the data. We will present the connectionist approach as technically equivalent to the latter process and due to its, alas, lack of any conclusive existent algorithm able to identify both the optimal model and architecture for a given problem, the three most common models presently at use and 20 different architectures for each model will be examined. It seems natural that if a comparison is to be made in order to provide a straight answer as to whether or not a connectionist approach outperforms the univariate Box&Jenkins methodology, the benchmark should clearly be the set of time series analysed in the work Time Series Analysis. Forecasting and Control by G. E. Box and G. M. Jenkins. Series BJA through to BJG give a total of 1200 plus measures to evaluate and compare the predictive power for different models, architectures, prediction horizons and pre-processing transformations.     

On the observer design in discrete-time

The paper deals with the equivalence, under coordinates change and output transformation, of a discrete-time nonlinear system to observer canonical forms. Necessary and sufficient conditions for local equivalence to these forms are given.     

Adaptive neural network control for a class of uncertain nonlinear systems in pure-feedback form

A procedure is developed for the design of adaptive neural network controller for a class of SISO uncertain nonlinear systems in pure-feedback form. The design procedure is a combination of adaptive backstepping and neural network based design techniques. It is shown that, under appropriate assumptions, the solution of the closed-loop system is uniformly ultimately bounded.     

On linear models for nonlinear systems

Best linear time-invariant (LTI) approximations are analysed for several interesting classes of discrete nonlinear time-invariant systems. These include nonlinear finite impulse response systems and a class of nonsmooth systems called bi-gain systems. The Fréchet derivative of a smooth nonlinear system is studied as a potential good LTI model candidate. The Fréchet derivative is determined for nonlinear finite memory systems and for a class of Wiener systems. Most of the concrete results are derived in an ?_∞ signal setting. Applications to linear controller design, to identification of linear models and to estimation of the size of the unmodelled dynamics are discussed.     

Multiple models switching control based on recurrent neural networks

This paper presents a novel approach in designing adaptive controller to improve the transient performance for a class of nonlinear discrete-time systems under different operating modes. The proposed scheme consists of generalized minimum variance (GMV) controllers and a compensating controller. GMV controllers are based on the known nominal linear multiple models, while the compensating controller is based upon a recurrent neural network. The adaptation law of network weight is derived from Lyapunov stability theory. A suitable switching control strategy is applied to choose the best controller by the performance indices at every sampling instant. Simulations are discussed in order to illustrate the merits of the proposed method.     

Computational intelligence approaches and linear models in case studies of forecasting exchange rates

Artificial neural networks and fuzzy systems, have gradually established themselves as a popular tool in approximating complicated nonlinear systems and time series forecasting. This paper investigates the hypothesis that the nonlinear mathematical models of multilayer perceptron and radial basis function neural networks and the Takagi–Sugeno (TS) fuzzy system are able to provide a more accurate out-of-sample forecast than the traditional auto regressive moving average (ARMA) and ARMA generalized auto regressive conditional heteroskedasticity (ARMA-GARCH) linear models. Using series of Brazilian exchange rate (R$/US$) returns with 15 min, 60 min and 120 min, daily and weekly basis, the one-step-ahead forecast performance is compared. Results indicate that forecast performance is strongly related to the series’ frequency and the forecasting evaluation shows that nonlinear models perform better than their linear counterparts. In the trade strategy based on forecasts, nonlinear models achieve higher returns when compared to a buy-and-hold strategy and to the linear models.     

A comparative evaluation of neural networks and hidden Markov models for monitoring turning tool wear

Condition monitoring of machine tool inserts is important for increasing the reliability and quality of machining operations. Various methods have been proposed for effective tool condition monitoring (TCM), and currently it is generally accepted that the indirect sensor-based approach is the best practical solution to reliable TCM. Furthermore, in recent years, neural networks (NNs) have been shown to model successfully, the complex relationships between input feature sets of sensor signals and tool wear data. NNs have several properties that make them ideal for effectively handling noisy and even incomplete data sets. There are several NN paradigms which can be combined to model static and dynamic systems. Another powerful method of modeling noisy dynamic systems is by using hidden Markov models (HMMs), which are commonly employed in modern speech-recognition systems. The use of HMMs for TCM was recently proposed in the literature. Though the results of these studies were quite promising, no comparative results of competing methods such as NNs are currently available. This paper is aimed at presenting a comparative evaluation of the performance of NNs and HMMs for a TCM application. The methods are employed on exactly the same data sets obtained from an industrial turning operation. The advantages and disadvantages of both methods are described, which will assist the condition-monitoring community to choose a modeling method for other applications.     

System identification of nonlinear state-space models

This paper is concerned with the parameter estimation of a general class of nonlinear dynamic systems in state-space form. More specifically, a Maximum Likelihood (ML) framework is employed and an Expectation Maximisation (EM) algorithm is derived to compute these ML estimates. The Expectation (E) step involves solving a nonlinear state estimation problem, where the smoothed estimates of the states are required. This problem lends itself perfectly to the particle smoother, which provides arbitrarily good estimates. The maximisation (M) step is solved using standard techniques from numerical optimisation theory. Simulation examples demonstrate the efficacy of our proposed solution.     

Identification of nonlinear dynamical systems using multilayered neural networks

A novel multilayer discrete-time neural net paradigm is presented for the identification of multi-input multi-output (MIMO) nonlinear dynamical systems. The major novelty of this approach is a rigorous proof of identification error convergence that reveals a requirement for a new identifier structure and nonstandard weight tuning algorithms. The NN identifier includes modified delta rule weight tuning and exhibits a learning-while-functioning feature instead of learning-then-functioning, so that the identification is on-line with no explicit off-line learning phase needed. The structure of the neural net (NN) identifier is derived using a passivity aproach. Linearity in the parameters is not required and certainty equivalence is not used. The notion of persistency of excitation (PE) and passivity properties of the multilayer NN are defined and used in the convergence analysis of both the identification error and the weight estimates.     

Nonlinear multivariable adaptive control using multiple models and neural networks

In this paper, a multivariable adaptive control approach is proposed for a class of unknown nonlinear multivariable discrete-time dynamical systems. By introducing a k-difference operator, the nonlinear terms of the system are not required to be globally bounded. The proposed adaptive control scheme is composed of a linear adaptive controller, a neural-network-based nonlinear adaptive controller and a switching mechanism. The linear controller can assure boundedness of the input and output signals, and the neural network nonlinear controller can improve performance of the system. By using the switching scheme between the linear and nonlinear controllers, it is demonstrated that improved performance and stability can be achieved simultaneously. Theory analysis and simulation results are presented to show the effectiveness of the proposed method.     

Artificial neural network models for indoor temperature prediction: investigations in two buildings

The problem how to identify prediction models of the indoor climate in buildings is discussed. Identification experiments have been carried out in two buildings and different models, such as linear ARX-, ARMAX- and BJ-models as well as non-linear artificial neural network models (ANN-models) of different orders, have been identified based on these experiments. In the models, many different input signals have been used, such as the outdoor and indoor temperature, heating power, wall temperatures, ventilation flow rate, time of day and sun radiation. For both buildings, it is shown that ANN-models give more accurate temperature predictions than linear models. For the first building, it is shown that a non-linear combination of sun radiation and time of day is important when predicting the indoor temperature. For the second building, it is shown that the indoor temperature is non-linearly dependent on the ventilation flow rate.     

Aspects of neural modeling of multidimensional chaotic attractors

This article discusses various aspects of neural modeling of multidimensional chaotic attractors. The Lorenz and Rosler attractors are considered as representative cases and are thoroughly examined. These two dynamical systems are expressed, within acceptable accuracy limits, by the corresponding systems of difference equations. Initially, the complete neural models of the attractors are examined. In this case, the neural networks are supplied with the values X_n , Y_n , Z_n of the systems to predict all the next components (X_{n +1}, Y_{n +1}, and Z_{n +1}) of the attractors. In the second case, named ‘component simulation’, the neural models are trained to predict only one of the values X_n , Y_n , Z_n , when they are fed with the complete input vector as in the first case. In the third case, the proposed neural networks are trained to predict only one component (X_{n +1}, Y_{n +1}, or Z_{n +1}) of the attractors, given a number of past values of the same component. Finally, the ability of the networks to predict the Y and Z components of an X time series of the dynamical systems is examined. Since the response of some networks is not satisfactory, the distribution of absolute error is considered in order to form a realistic picture of the networks’ performance.