Journal of Zhejiang University SCIENCE C

Echo state network (ESN), which efficiently models nonlinear dynamic systems, has been proposed as a special form of recurrent neural network. However, most of the proposed ESNs consist of complex reservoir structures, leading to excessive computational cost. Recently, minimum complexity ESNs were proposed and proved to exhibit high performance and low computational cost. In this paper, we propose a simple deterministic ESN with a loop reservoir, i.e., an ESN with an adjacent-feedback loop reservoir. The novel reservoir is constructed by introducing regular adjacent feedback based on the simplest loop reservoir. Only a single free parameter is tuned, which considerably simplifies the ESN construction. The combination of a simplified reservoir and fewer free parameters provides superior prediction performance. In the benchmark datasets and real-world tasks, our scheme obtains higher prediction accuracy with relatively low complexity, compared to the classic ESN and the minimum complexity ESN. Furthermore, we prove that all the linear ESNs with the simplest loop reservoir possess the same memory capacity, arbitrarily converging to the optimal value.

This study deals with the neuro-fuzzy (NF) modelling of a real industrial winding process in which the acquired NF model can be exploited to improve control performance and achieve a robust fault-tolerant system. A new simulator model is proposed for a winding process using non-linear identification based on a recurrent local linear neuro-fuzzy (RLLNF) network trained by local linear model tree (LOLIMOT), which is an incremental tree-based learning algorithm. The proposed NF models are compared with other known intelligent identifiers, namely multilayer perceptron (MLP) and radial basis function (RBF). Comparison of our proposed non-linear models and associated models obtained through the least square error (LSE) technique (the optimal modelling method for linear systems) confirms that the winding process is a non-linear system. Experimental results show the effectiveness of our proposed NF modelling approach.

We present how residual intensity modulation (RIM) affects the performance of a resonator fiber optic gyro (R-FOG) through a sinusoidal wave phase modulation technique. The expression for the R-FOG system’s demodulation curve under RIM is obtained. Through numerical simulation with different RIM coefficients and modulation frequencies, we find that a zero deviation is induced by the RIM effect on the demodulation curve, and this zero deviation varies with the RIM coefficient and modulation frequency. The expression for the system error due to this zero deviation is derived. Simulation results show that the RIM-induced error varies with the RIM coefficient and modulation frequency. There also exists optimum values for the RIM coefficient and modulation frequency to totally eliminate the RIM-induced error, and the error increases as the RIM coefficient or modulation frequency deviates from its optimum value; however, in practical situations, these two parameters would not be exactly fixed but fluctuate from their respective optimum values, and a large system error is induced even if there exists a very small deviation of these two critical parameters from their optimum values. Simulation results indicate that the RIM-induced error should be considered when designing and evaluating an R-FOG system.

The shapes of speakers’ vocal organs change under their different emotional states, which leads to the deviation of the emotional acoustic space of short-time features from the neutral acoustic space and thereby the degradation of the speaker recognition performance. Features deviating greatly from the neutral acoustic space are considered as mismatched features, and they negatively affect speaker recognition systems. Emotion variation produces different feature deformations for different phonemes, so it is reasonable to build a finer model to detect mismatched features under each phoneme. However, given the difficulty of phoneme recognition, three sorts of acoustic class recognition—phoneme classes, Gaussian mixture model (GMM) tokenizer, and probabilistic GMM tokenizer—are proposed to replace phoneme recognition. We propose feature pruning and feature regulation methods to process the mismatched features to improve speaker recognition performance. As for the feature regulation method, a strategy of maximizing the between-class distance and minimizing the within-class distance is adopted to train the transformation matrix to regulate the mismatched features. Experiments conducted on the Mandarin affective speech corpus (MASC) show that our feature pruning and feature regulation methods increase the identification rate (IR) by 3.64% and 6.77%, compared with the baseline GMM-UBM (universal background model) algorithm. Also, corresponding IR increases of 2.09% and 3.32% can be obtained with our methods when applied to the state-of-the-art algorithm i-vector.

We propose novel techniques to find the optimal location, size, and power factor of distributed generation (DG) to achieve the maximum loss reduction for distribution networks. Determining the optimal DG location and size is achieved simultaneously using the energy loss curves technique for a pre-selected power factor that gives the best DG operation. Based on the network’s total load demand, four DG sizes are selected. They are used to form energy loss curves for each bus and then for determining the optimal DG options. The study shows that by defining the energy loss minimization as the objective function, the time-varying load demand significantly affects the sizing of DG resources in distribution networks, whereas consideration of power loss as the objective function leads to inconsistent interpretation of loss reduction and other calculations. The devised technique was tested on two test distribution systems of varying size and complexity and validated by comparison with the exhaustive iterative method (EIM) and recently published results. Results showed that the proposed technique can provide an optimal solution with less computation.

Data sparseness, the evident characteristic of short text, has always been regarded as the main cause of the low accuracy in the classification of short texts using statistical methods. Intensive research has been conducted in this area during the past decade. However, most researchers failed to notice that ignoring the semantic importance of certain feature terms might also contribute to low classification accuracy. In this paper we present a new method to tackle the problem by building a strong feature thesaurus (SFT) based on latent Dirichlet allocation (LDA) and information gain (IG) models. By giving larger weights to feature terms in SFT, the classification accuracy can be improved. Specifically, our method appeared to be more effective with more detailed classification. Experiments in two short text datasets demonstrate that our approach achieved improvement compared with the state-of-the-art methods including support vector machine (SVM) and Naïve Bayes Multinomial.