This work uses the weights and network output of neural networks (NN) as state equation and measurement equation for chaotic time-series' prediction to obtain the linear state transition equation which is different from the previous filtering methods based chaotic time-series' prediction, and the prediction results of chaotic time series is represented by the predicted measurement value. An efficient algorithm with continuous update prediction scheme for chaotic time-series is suggested. This scheme is tested using simulated data based on the EKF, UKF and Gaussian particle filtering (GPF), respectively. Simulation results have proved that the GPF is superior to EKF and UKF for Mackey-Glass time-series' prediction with the proposed model proposed in this paper.

This study introduces a research path to obtain alternative trading rules by using nonlinear dynamical analysis of stock returns. We examine the daily return data of Istanbul Stock Exchange index and Shenzhen Index B-Shares. Both stock returns series are shown to exhibit chaotic behavior and associated maximal Lyapunov exponents (LE) are computed. A new prediction method which bases on the properties of detected chaotic behavior is proposed to perform one-week out-of-sample prediction of the stock returns. Finally we develop a nonlinear model of active trading, in which traders rely only on their heterogeneous forecasts of future periods' maximum and minimum returns. The model motivates active trading under chaotic behavior.

This paper studies the dynamics and stability of a stage-structured (immature/mature) population. The emphasis is on how resource available to the immature stage indirectly regulates the dynamics of the mature stage, where the resource is defined by a Holling-II function. We present the results from numerical experiments, where inference about stability and predictability of the system dynamics is based on the theory of Lyapunov exponents.

This paper proposes the adaptive prediction model of software failure data analysis based on Volterra Series combined the theory of phase space reconstruction. The model makes full use of the advantages of small computational complexity, fast convergence speed and strong adaptive ability of Volterra series, to well reflect the dynamic characteristics of the software failure system. Numerical simulation results of practical failure data series show that Volterra model improves the prediction precision of traditional chao local model, also possesses a strong the anti-noise performance and can hold an optimistic prediction effect even if the data series have been polluted by the white noise.

This chapter provides an overview of E-nose research and technology. We first review the progress of E-noses in applications, systems, and algorithms during the past two decades. Then, we propose to address these key challenges in E-nose, which are sensor induced and sensor specific. This chapter is closed by a statement of the objective of the research, a brief summary of the work, and a general outline of the overall structure of this book.

There are many practical situations in which the chaotic signal appears in the observation in a random manner so that there are intermittent failures in the observation mechanism at certain times. Using weights and network output of neural network as state equation and observation equation to obtain the linear state transition equation, and the chaotic time-series prediction results represented by the predicted observation value, this paper generalizes the extended Kalman filter (EKF) to the case for the prediction of chaotic time-series with intermittent observations when random interruptions in the observation process are modeled by a sequence of independent Bernoulli random variables. Finally, we test this scheme using simulated data based on the generalized EKF with different Bernoulli distribution probability for uncertain observations. Simulation results of Lorenz time-series prediction with synthetic data prove that the proposed algorithm in this paper has satisfactory prediction precision as well as good robustness.

The prediction of chaotic time series is performed by relevance vector machine (RVM), which is an inherent online machine learning technique utilizing a flexible and sparse function without additional regularization parameters. The main objective of this approach is to increase the accuracy of the chaotic time series prediction. The method is applied to Mackey-Glass and Lorenz equations, Henon mapping which produce the chaotic time series to evaluate the validity of the proposed technique. Numerical experimental results confirm that the proposed method can predict the chaotic time series more effectively and accurately when compared with the existing prediction methods.

The prediction of chaotic time series is an important research issue. To improve the prediction accuracy, a hybrid approach called WNN-PSO is proposed, which based on the self-learning ability of wavelet neural network, whose parameters are optimized by particle swarm optimization. The WNN-PSO method has higher prediction accuracy, fast convergence, and heightens the ability of jumping the local optimums. The experiment results of the prediction for chaotic time series show the feasibility and effectiveness of the proposed method. Compared with wavelet neural network and BP neural network, the proposed method are superior to them. Finally, the WNN-PSO is applied to predict the life energy consumption of china in our lives.

The enthusiasm of deep analysis of nonlinear dynamical system turns on the idea of chaos. Models of chaotic system have been used to analyze nonlinear dynamics in speech signal which has been traditionally modeled as a linear system for several decades. Nonlinear modeling of speech opens a new horizon in the field of cochlear implant devices in various aspects. These devices are designed to provide a measure of hearing to the deaf. It has been very successful in restoring partial hearing to profoundly deaf people. Despite all benefits one of the major issues is a great variability in the speech-recognition performance of cochlear implant patient. For a given type of implant, auditory performance which is defined as the ability to discriminate, detect, identify or recognize speech may vary from 0 to 100% The factors which affect are lack of understanding about cognitive and perceptual mechanism of individuals, numbers of nerve fibers and processing of sensory information. Neural response telemetry is a technique applied for testing of device to find clue for onepsilas actual hearing perception. In this technique evoked action potential has been measured which provides estimation about loudness levels of patient in different position of electrodes. This estimate does not tell about formants missing and frequency information. In this paper, chaotic analysis of speech has been presented for comparison of normal hearing response and implanted patient response.

The article presents the possibility of applying selected tools of chaos theory in the process of modeling a frictional pair by monitoring its operating condition aspects. These tools are suggested with regard to the friction process, which subtracts a variety of noises currently treated as a stochastic phenomenon. This assumption significantly reduces the possibility of a precise description and modeling of friction phenomena. The fundamental assumption that enables the application of chaos theory tools is that of treating the friction phenomenon as a non-linear dynamic one. This assumption allows for an analysis of the basic values that characterize the operation of a frictional pair in phase space. Such a manner of presenting and analyzing the parameters of the operation of a frictional pair allows for a prediction of the system’s condition (the parameters describing the operating conditions of the frictional pair) in a short period of time. In this article, the authors implemented the above-mentioned method to the modeling of an automotive disk brake couple. This tribo-system was selected because of the great intensiveness of the friction process and the non-trivial impact of noises. The article proves that the friction process, according to the performed tests, comes under chaos theory properties. Moreover, a successful prediction of the operating conditions of the frictional pair was carried out.

Prospective infectious disease outbreak detection has long been a major concern in public health. Using time series analysis method for the outbreak detection, a nonlinear Markov switching model is more excellent than linear models in modelling time series, due to its ability to describe the switching process of time series variables in different states. However the estimation difficulty of Markov switching model hinders the model's extensive application in practice. The paper proposes using Differential Evolution (termed DE) algorithm to obtain maximum likelihood estimator of Markov switching model in consideration of DE's good global optimization ability. In addition, to effectively reduce negative impact of label switching problem on disease outbreak detection validity of the estimated model by maximum likelihood estimation (termed MLE) method, the paper introduces identifiability constraint on estimation parameters constructed with the heuristic information about difference between durations of different states in MLE using DE. Encouraging experimental study has demonstrated the effectiveness and efficiency of DE in maximizing likelihood function of the studied Markov switching model as well as the effectiveness of the proposed identifiability constraint on improving disease outbreak detection validity of the estimated Markov switching model by MLE.

Presetting control parameters of algorithms are important to ant colony optimization (ACO). This paper presents an investigation into the relationship of algorithms performance and the different control parameter settings. Two tour building methods are used in this paper including the max probability selection and the roulette wheel selection. Four parameters are used, which are two control parameters of transition probability α andβ, pheromone decrease factor ρ, and proportion factor q0 in building methods. By simulated result analysis, the parameter selection rule will be given.

On the assumption that random interruptions in the observation process are modelled by a sequence of independent Bernoulli random variables, this paper generalize the extended Kalman filtering (EKF), the unscented Kalman filtering (UKF) and the Gaussian particle filtering (GPF) to the case in which there is a positive probability that the observation in each time consists of noise alone and does not contain the chaotic signal (These generalized novel algorithms are referred to as GEKF, GUKF and GGPF correspondingly in this paper). Using weights and network output of neural networks to constitute state equation and observation equation for chaotic time-series prediction to obtain the linear system state transition equation with continuous update scheme in an online fashion, and the prediction results of chaotic time series represented by the predicted observation value, these proposed novel algorithms are applied to the prediction of Mackey–Glass time-series with additive and multiplicative noises. Simulation results prove that the GGPF provides a relatively better prediction performance in comparison with GEKF and GUKF.

LEARNING LATENT TEMPORAL MANIFOLDS FOR RECOGNITION AND PREDICTION OF MULTIPLE ACTIONS IN STREAMING VIDEOS USING DEEP NETWORKS Name: Nair, Binu Muraleedharan University of Dayton Advisor: Dr. Vijayan K. Asari Recognizing multiple types of actions appearing in a continuous temporal order from a streaming video is the key to many possible applications ranging from real-time surveillance to egocentric motion for human computer interaction. Current state of the art algorithms are more focused either on holistic video representation or on finding a specific activity in video sequences. But the major drawback is that these algorithms work only on applications pertaining to unconstrained video search from the web and requires the complete sequence for reporting what kind of actions are present. In this dissertation, we propose an algorithm to detect and recognize multiple actions in a streaming sequence at every instant. This approach was successful in recognizing the type of action being performed and also provides a percentage of completion of that action at every instant in realtime. This system is invariant to the number of frames and the speed at which the action is being performed. Apart from these benefits, the proposed model can also predict the motion descriptors at future instances corresponding to the action present. Since human motion is inherently continuous in nature, the algorithm presented in this dissertation computes novel motion descriptors based iii on the dense optical flow at every instant and evaluates their variations along the temporal domain using deep learning techniques. For each action type, we compute a non-linear transformation from motion descriptor space into the latent temporal space using stacked autoencoders where this transformation is learned from its training patterns. The latent features thus obtained, forms a temporal manifold where the transitions along it are modeled using the conditional restricted Boltzmann machines (CRBMs). Using these trained autoencoders and CRBMs for every action type, we can make an inference into multiple latent temporal action manifolds at an instant from a set of streaming input frames. Our model achieved a high accuracy of 93% in recognizing actions per frame and was able to predict the future action instances with an accuracy of 84% for KTH dataset. Similarly, it was also tested with the UCF Sports dataset and achieved an accuracy of 84% in recognizing the action per-frame and around 69% of predictive capability. Therefore we believe that the proposed model can benefit applications in human computer interaction, gaming and IP surveillance where the action classification using temporal manifolds and its predictive capability are crucial.

In this paper we firstly analysis the chaotic characters of three sets of the financial time series (Hang Sheng Index (HIS), Shanghai Stock Index and US gold price) based on the phase space reconstruction. But when we adopt the feedforward neural networks to predict those time series, we found this method run short of a criterion in selecting the training set, so we present a new method: using correlation dimension (CD) as the criterion. By the experiments, the method is proved effective.

In order to solve the problem of rapid recognizing chaotic waves in the real-time, according to the characteristics of chaotic waves, a new method is proposed. The method can avoid a lot more calculating work, and simplifies measuring chaotic waves to achieve fast preliminary recognition efforts. Based on Chua’s circuit, the mechanism of the method is introduced in detail. Periodic signal, noise signal, single-scroll chaotic and double scroll chaotic waves and other waves have been experimented physically, which verify that the method is simple and effective. All figures are measured by a digital oscilloscope in the paper.

In order to overcome the shortcomings (such as the time–frequency localization and the nonstationary signal analysis ability) of the Fourier transform, time–frequency analysis has been carried out by wavelet packet decomposition and reconstruction according to the actual nonstationary vibration signal from a large equipment located in a large Steel Corporation in this article. The effect of wavelet decomposition on signal denoising and the selection of high-frequency weight coefficients for each layer on signal denoising were analyzed. The nonlinear prediction of the chaotic time series was made by global method, local method, weighted first-order local method, and maximum Lyapunov exponent prediction method correspondingly. It was found the multi-step prediction method is better than other prediction methods.

Generally the neural network and the fuzzy compensative algorithm are applied to forecast the time series for power demand with a characteristic of non-linear dynamic system, but it has a few prediction errors relatively. It also makes long term forecast difficult for sensitivity on the initial condition. On this paper, we evaluate the chaotic characteristic of electrical power demand with analysis methods of qualitative and quantitative and perform a forecast simulation of electrical power demand in regular sequence, attractor reconstruction, time series forecast for multi dimension using Lyapunov exponent quantitatively. We compare simulated results with the previous method and verify that the purpose one being more practice and effective than it.

From the viewpoint of nonlinear dynamics, a numerical method for predicting the traffic of the local area network (LAN) is presented based on the multifractal spectrums; in particular, for predicting the typical congestion and bursting phenomena, by analyzing real time sequences. First, the multifractal spectrums available to the LAN traffic are derived in some detail and their physical meanings are consequently explained. Then an exponent factor is introduced to the measurement or description of the singularity of the time sequence and the correlations between multifractal spectrums and traffic flow rate are studied in depth. Finally, as an example, the multifractal spectrums presented are used to predict the network traffic of an Ethernet by analyzing its real time sequence. The results show that there exists a distinct relationship between the multifractal spectrums and the traffic flow rate of networks and the multifractal spectrum could be used to efficiently and feasibly predict the traffic flow rate, especially for predicting the singularities of the real time sequences, which are closely related to the congestion and bursting phenomena. Thus, this method can be applied to the prediction and management of the congestion and bursting in the network traffic at an early time. Furthermore, the prediction will become much more accurate and powerful over a long period, since the fluctuations of the traffic flow rate are remarkable.

For many engineering and aerospace applications, detection and quantification of multiscale damage in fiber-reinforced composite structures is increasing in importance. Consequently, the development of an efficient and cost-effective diagnosis scheme that can accurately sense, characterize, and evaluate the existence of any form of damage will offer significant potential for improving the performance, reliability, and extending the operational life of these complex systems. We present an approach to characterize and classify different damage states in composite laminates by measuring the change in the signature of the resultant wave that propagates through the anisotropic media under forced excitation. The wave propagation is measured using surface-mounted piezoelectric transducers. Sensor signals collected from test specimens with various forms of induced damage are analyzed using a pattern-recognition algorithm known as the one-class support vector machines. The one-class support-vector-machine algorithm performs automatic anomaly detection and classification of damage signatures using various features from the sensor readings. The results obtained suggest that the one-class support-vector-machine algorithm, along with appropriate preprocessing techniques, can often achieve better accuracy than the popular k-nearest-neighbor method in detecting and classifying anomalies caused by structural defects, even when the perturbations caused in the sensor signals due to different damage states are minimal.