Nonlinear system identification using optimized dynamic neural network

In this paper, both off-line architecture optimization and on-line adaptation have been developed for a dynamic neural network (DNN) in nonlinear system identification. In the off-line architecture optimization, a new effective encoding scheme—Direct Matrix Mapping Encoding (DMME) method is proposed to represent the structure of neural network by establishing connection matrices. A series of GA operations are applied to the connection matrices to find the optimal number of neurons on each hidden layer and interconnection between two neighboring layers of DNN. The hybrid training is adopted to evolve the architecture, and to tune the weights and input delays of DNN by combining GA with the modified adaptation laws. The modified adaptation laws are subsequently used to tune the input time delays, weights and linear parameters in the optimized DNN-based model in on-line nonlinear system identification. The effectiveness of the architecture optimization and adaptation is extensively tested by means of two nonlinear system identification examples.     

Development of piezoelectric fans for flapping wing application

A piezoelectric fan (piezofan) which couples a piezoelectric unimorph to an attached flexible blade is able to produce a large deflection especially at resonance. The fundamental resonant frequencies (f_r) of the piezofan structures have been calculated by an analytical method and finite element modelling, and these were compared with experimental measurements. Good agreements have been obtained between them. The free tip deflection at quasi-static operation or/and the vibration amplitude at dynamic operation (A) of the piezofans have been experimentally measured. We introduce f_r × A as an optimization criterion for piezofans. Optimization according to this criterion has been carried out for some piezofan configurations, such as the length and the location of the piezo patch, as well as the thickness ratio between the elastic and piezoelectric layers among a few available variations. Results show this optimization approach to be promising when compared to previously defined piezofan performance parameters such as the energy transmission coefficient and electromechanical coupling coefficient.     

Clonal optimization-based negative selection algorithm with applications in motor fault detection

The Negative Selection Algorithm (NSA) and clonal selection method are two typical kinds of artificial immune systems. In this paper, we first introduce their underlying inspirations and working principles. It is well known that the regular NSA detectors are not guaranteed to always occupy the maximal coverage of the nonself space. Therefore, we next employ the clonal optimization method to optimize these detectors so that the best anomaly detection performance can be achieved. A new motor fault detection scheme using the proposed NSA is also presented and discussed. We demonstrate the efficiency of our approach with an interesting example of motor bearings fault detection, in which the detection rates of three bearings faults are significantly improved.     

Efficient simulation of hard disk drive operational shock response using model order reduction

The complex structure, coupled mechanical and fluidic energy domains, and inherent nonlinearity of air bearing between slider and disk involved in the hard disk drive (HDD) are normally presented as a large scale problem which will result in very heavy computational costs in terms of intensive computation and time consuming for HDD research communities and industries to carry out the transient dynamic simulation for HDD design verification, performance analysis, and optimization by using the traditional full-order models, such as finite element model (FEM). This paper presents a method of application of model order reduction (MOR) technique to dramatically reduce the computation time for HDD transient shock performance analysis while capturing the behaviors of original problem faithfully. The reduced models are obtained by performing MOR directly to the FEMs through Krylov subspace and Arnoldi algorithm. The transient operational shock response results of the reduced models of a head suspension assembly (HSA) subjected to half-sine shock pulse demonstrate that the reduced models can dramatically reduce total computation by at least three orders and have very good agreement with those simulated from the original large problem by full-order FEM.     

Identification of relationship between sunspots and natural runoff in the Yellow River based on discrete wavelet analysis

Annual natural runoff is an important index of a river, which may be affected by solar activities. In this study, 304 years of annual natural runoff at the Sanmenxia station located in the Yellow River and the sunspot relative number are decomposed with the application of a Complex Morlet. According to the results of real part, modulus and second power of modulus, the annual runoff series at the Sanmenxia station has an obvious periodic oscillation on 90–100, 50–80, 35–50, 15–35, about 10, and less than 10-year scales. Also, there are obvious periodic variability with 60–90 years, 30–50 years and about 10 years. There are two centers of energy: one is about 1840–1850 on 7–11-year scale and the other is about 1825–1925 on 60–70-year scale. From the wavelet variance, 3, 26, 46, 68 year periods are detected within a 100-year scale, and the 68-year period is the most significant. Similar analyses are conducted for the sunspot relative number within the same period 1700–2003. The sunspot series shows 11- and 60-year period variation, as well as eight energy centers. Then, the correlation analyses for 11- and 60-year serial scales are computed. From a long-term period (1700–2003) view, there is no notable correlation between the natural runoff and the sunspot relative number; however, it is evident that the correlations exist within a short-term period. The results also indicate that the relationships between solar activities and the natural runoff in the Yellow River are complicated.     

基于负载高峰特征的虚拟机放置算法

提出了一种基于虚拟机负载高峰特征的虚拟机放置策略,通过更好地复用物理主机资源来实现资源共享,从而提高资源利用率.在云环境下,当多个虚拟机的负载高峰出现在相同的时间段内时,非高峰时段的资源利用率就会明显偏低;相反,多个虚拟机只要负载高峰能错开在不同的时间,闲置的资源就能更充分地被利用.由于应用的负载通常具有一定的周期性,因此,可以利用虚拟机负载的历史数据作为分析的依据.基于虚拟机的负载高峰特征对虚拟机负载进行建模,建立虚拟机负载之间的相似度矩阵来实现虚拟机联合放置.使用CloudSim模拟实现了所提出的算法,并与基于相关系数的放置算法、随机放置算法进行了比较.实验结果表明:所提算法在平均CPU利用率上有8.9%~12.4%的提高,主机使用量有8.2%~11.0%的节省.     

Analysis of models of the integral cellular communication network with different schemes of channel partitioning

Two schemes of partitioning the shared channel pool between the calls of speech and data in the integral cellular communication networks were proposed. One uses the isolated partitioning procedure, the other, the virtual partitioning. The isolated partitioning forbids channel reassignment from one zone to another, whereas the virtual partitioning relies exactly on this procedure. In both schemes, the voice calls first occupy the free channels in their zone, and only if the channels of this zone are occupied, then a free channel is sought in another zone only for the handover voice call. In this case, the maximal number of handover speech calls in the zone of channels for data calls is limited. A scheme constraining the access of such calls depending on the total number of data calls in the cell is used to establish a rule for access of new data calls to the channels. Methods were developed to calculate the indices of service performance with the proposed access strategies. The results of numerical experiments were presented, and for different access strategies the indices were compared.     

A new phase unwrapping method based on region recognition and region expansion

Phase unwrapping is a key step in retrieving digital elevation models (DEMs) from across-track interferometric synthetic aperture radar (InSAR) data. The coherence of synthetic aperture radar (SAR) data set is an effective indicator for the quality of phase unwrapping. However, the coherence of different regions usually distributes unevenly in SAR images monitoring heterogeneous areas. Errors in low-coherence areas are prone to pollute the whole image. In order to mitigate propagation error, a new phase unwrapping algorithm based on region recognition and region expansion is proposed. In the region recognition step, optical images are incorporated to recognize low-coherence regions by virtue of supervised classification technique. Low-coherence regions and the ones that are not of interest for the application are then discarded. In the region expansion step, stable pixels of high coherence are selected as growing seeds, and then phase unwrapping grows from high-quality regions to low-quality ones guided by coherence information and weighted numbers of neighbouring unwrapped pixels. The ambiguity number of a wrapped pixel is estimated from its neighbouring pixels under the criteria of pixel distance and phase gradient. Iterative examination continues until the whole image is unwrapped. Experiments on PALSAR and ASAR data demonstrate its validity and advantages over other classical methods.     

An efficient method based on machine learning for estimation of the wall parameters in through-the-wall imaging

The estimation of the wall parameters is important in through-the-wall radar imaging (TWRI). Ambiguities in the wall characteristics, including wall thickness, permittivity, and conductivity, will distort the imaging and shift the target position. To obtain a quick and accurate estimation of wall parameters, an efficient method based on machine learning is proposed. The estimation problem is converted to a regression problem. A map between wall parameters and the received signals is established and is regressed as a linear formulation after machine learning; in this manner, the wall parameters can be estimated in few seconds. The measurement results demonstrate that the estimated approach has the advantages of high precision and low computational time. The influence of the size, the location, the number of the targets and the length of the wall, the sampling interval, and noise on the estimation problems is discussed, and the image entropy is given to verify the effectiveness of the estimation values. The results based on support vector machines and least-square support vector machines (LS-SVMs), which are both machine-learning approaches, are compared. The comparison results reveal that the LS-SVM-based method can provide comparable performances in terms of accuracy and convenience but poor performances in terms of generalization and robustness.