From batch to recursive outlier-robust identification of non-linear dynamic systems with neural networks

IFAC Proceedings Volumes

2009

Simulation is useful for the evaluation of a Master Production/distribution Schedule (MPS). Also, the goal of this paper is the study of the design of a simulation model by reducing its complexity. According to theory of constraints, we want to build reduced models composed exclusively by bottlenecks and a neural network. Particularly a multilayer perceptron, is used. The structure of the network is determined by using a pruning procedure. This work focuses on the impact of discrete data on the results and compares different approaches to deal with these data. This approach is applied to sawmill internal supply chain.

“…-Learning of the parameters by using Levenberg-Marquard algorithm with robust criterion (Thomas and Bloch 1996).…”

Section: The Multilayer Perceptronmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

How deals with discrete data for the reduction of simulation models using neural network

IFAC Proceedings Volumes

2009

“…-Learning of the parameters by using LevenbergMarquard algorithm with robust criterion [24]. -Weights elimination by using the Optimal Brain Surgeon algorithm with a robust criterion [25].…”

Section: The Multilayer Perceptron (Mlp)mentioning

confidence: 99%

Simulation Reduction Models Approach Using Neural Network

Choffel

Tenth International Conference on Computer Modeling and Simulation (Uksim 2008)

2008

Adaptive Control & Signal

“…An outlier may be defined as a sample in a time series, which is inconsistent with the rest of the samples [15]. These samples are often difficult to eliminate [16]. A typical time series usually contains around 1% to 10% of such outlier samples [17], which may be introduced 387 2.…”

Section: Introductionmentioning

confidence: 99%

Development of a novel robust identification scheme for nonlinear dynamic systems

George

Panda

2014

This paper presents a set of single layer low complexity nonlinear adaptive models for efficient identification of dynamic systems in the presence of outliers in the training signal. The weights of the new models have been updated using a new robust learning algorithm. The proposed robust algorithm is based on adaptive minimization of Wilcoxon norm of errors. The computational complexity associated with the new models has further been reduced by processing the input in block form and using a newly derived robust block learning algorithm. Through exhaustive simulation study of many benchmark identification examples, it has been shown that in all cases, the new models provide enhanced and robust identification performance compared with that provided by the corresponding conventional squared error-based approaches. Copyright by temporary or permanent breakdown of sensors, analog-to-digital conversion errors, failure of transducers, and so on.In the recent past, many robust learning algorithms, which are efficient in learning in the presence of outliers, have been proposed in the literature. A robust recursive least square method has been developed for identification of bilinear systems [18]. Another robust learning algorithm, which employs a soft filtering approach to remove the most prominent outliers before parameter estimation, has been derived in [19]. The technique has been applied to develop a robust recurrent neural network. Sanchez [20] has proposed a robust learning method for RBF networks, in which an RBF architecture is generated in the first phase using conventional learning scheme, and subsequently, a robust learning method is introduced to make the network robust against outliers. The mechanism by which outliers affect neural network models has been studied in [21] using the analysis of its influence function. The author also suggested the need for considering the use of error distribution functions with extensive tails, like the Cauchy or Lorentzian distribution. A mean log squared error cost function has also been proposed, which is more robust compared with conventional least mean square approach. Other prominent robust learning techniques include random sample consensus method [22] and M -estimator sampling consensus method [23].By combining conventional robust techniques with optimization of learning rate, a robust learning algorithm for optimizing fuzzy neural networks has been suggested in [24]. A robust fuzzy regression agglomeration clustering algorithm has been presented in [25] to develop Takagi-Sugeno-Kang models, which are robust against outliers. They have tested the performance through simulation studies. A hybrid robust Takagi-Sugeno-Kang modeling approach has been recently proposed in [26] by using a robust fuzzy C-regression model clustering algorithm in the coarse-tuning stage and an annealing robust back propagation learning algorithm in the fine-tuning phase. A robust support vector regression network has been proposed by employing traditional robust learning approaches to improve ...