Search

This paper proposes an adaptive ensemble learning strategy for soft sensor development with semi-supervised learning. The main target of the proposed method is to improve the regression performance with a limited number of labeled samples, under the ensemble learning framework. First, the missing outputs are estimated by the k -nearest neighbor method. In order to improve the accuracy of sub-models for ensemble modeling, a novel sample selection mechanism is established to select the most useful estimated data samples. Second, the Bagging method is employed to both of the labeled and selected datasets, and the two different kinds of datasets are matched based on the Dissimilarity algorithm. As a result, the proposed method enhances the diversity and accuracy of the sub-models which are two important issues for ensemble learning. An industrial case study is carried out to demonstrate the effectiveness of the proposed method in dealing with semi-supervised soft sensing issue. [ABSTRACT FROM AUTHOR]

Considering the deviation of the working condition and the high updating frequency of the traditional moving window methods, this paper proposes a selective strategy of moving window for the Gaussian process regression in the latent probabilistic component space. First, the probabilistic principle component analysis (PPCA) is employed to deal with the multi-dimensional issue and extract essential information of the process data. Because the latent probabilistic components are more sensitive to the deviation of the working condition in the industrial process than the original data, the regression performance is improved under the PPCA framework. Under the proposed strategy, the soft sensor is able to detect the change of the working condition, and the updating is activated only when the predicted error exceeds the preset threshold, otherwise the model is kept unchanged. Furthermore, the promotion of both predicted accuracy and efficiency can be obtained by regulating the threshold. To test the effectiveness of the proposed method, a wastewater case study is provided, and the result shows that the proposed strategy works better under the probabilistic than other conventional methods. [ABSTRACT FROM AUTHOR]

This paper considers the parameter estimation for Wiener time-delay systems with the output data contaminated with outliers. The time-delay and corrupted output data bring great challenges to the parameter estimation problem. The statistical model of the estimation problem is constructed based on the Laplace distribution and the identification problem is formulated in the scheme of the expectation-maximization (EM) algorithm. The negative effect of outliers imposed on the parameter estimation problem is sufficiently suppressed and the unknown time-delay and model parameters can be estimated simultaneously. The simulation example is given to demonstrate the effectiveness of the proposed algorithm. [ABSTRACT FROM AUTHOR]

This paper considers the parameter estimation problem for Hammerstein systems with saturation and preload nonlinearities. Based on the key term separation technique, the output of the system is expressed as a linear combination of all the system parameters. By introducing the forgetting factors and using the data filtering technique, a data filtering based forgetting factor stochastic gradient (F-FF-SG) algorithm is presented. The simulation examples illustrate that the F-FF-SG algorithm has faster convergence rates and better parameter estimation accuracies than the stochastic gradient algorithm and the data filtering based stochastic gradient algorithm. [ABSTRACT FROM AUTHOR]

Abstract: This paper considers the problem of robust stability of Cohen–Grossberg neural networks with time-varying delays. Based on the Lyapunov stability theory and linear matrix inequality (LMI) technique, some sufficient conditions are derived to ensure the global robust convergence of the equilibrium point. The proposed LMI conditions can be checked easily by recently developed algorithms solving LMIs. Comparisons between our results and previous results admits our results establish a new set of stability criteria for delayed Cohen–Grossberg neural networks. Numerical examples are given to illustrate the effectiveness of our results. [Copyright &y& Elsevier]

Abstract Robust identification of the linear parameter varying (LPV) finite impulse response (FIR) model with time-varying time delays is considered in this paper. A robust observation model based on Laplace distribution is established to deal with the output data contaminated with the outliers, which are commonly existed in modern industries. A Markov chain model is utilized to model the correlation between the time delays as they do not simply change randomly in reality. A transition probability matrix and an initial probability distribution vector are used to govern the switching mechanism of the time delays. Since it is difficult to optimize the complex log likelihood function directly, the derivations of the proposed algorithm are performed under the framework of Expectation-Maximization (EM) algorithm. A numerical example and a chemical process are utilized to verify the effectiveness of the proposed approach. [ABSTRACT FROM AUTHOR]

This paper puts forward a robust identification solution for nonlinear time-delay state-space model (NDSSM) with contaminated measurements. To enhance the robustness of the developed method for outliers, the heavy-tailed Laplace distribution is employed to describe and protect the output measurement process. The undetermined time-delay is considered to be uniformly distributed and the boundary of it is known as a priori. In the developed solution, the uncertain time-delay is treated as a latent process variable and it is iteratively calculated with the expectation–maximization (EM) algorithm. The EM algorithm is actually an iterative optimization algorithm and it is effective for the hidden variable problems. The particle filter is introduced to numerically approximate the cost function (Q-function) in the EM algorithm since it is difficult to calculate directly. The efficacy of the developed solution is evaluated via a numerical test and a two-link robotic manipulator. [ABSTRACT FROM AUTHOR]