Home About us Contact
|
Home About us Contact | ||
|
|
||
Local Optimal Solutions (local + optimal_solution)
Selected AbstractsMemetic evolutionary training for recurrent neural networks: an application to time-series predictionEXPERT SYSTEMS, Issue 2 2006M. Delgado Abstract: Artificial neural networks are bio-inspired mathematical models that have been widely used to solve complex problems. The training of a neural network is an important issue to deal with, since traditional gradient-based algorithms become easily trapped in local optimal solutions, therefore increasing the time taken in the experimental step. This problem is greater in recurrent neural networks, where the gradient propagation across the recurrence makes the training difficult for long-term dependences. On the other hand, evolutionary algorithms are search and optimization techniques which have been proved to solve many problems effectively. In the case of recurrent neural networks, the training using evolutionary algorithms has provided promising results. In this work, we propose two hybrid evolutionary algorithms as an alternative to improve the training of dynamic recurrent neural networks. The experimental section makes a comparative study of the algorithms proposed, to train Elman recurrent neural networks in time-series prediction problems. [source] Satellite image segmentation using hybrid variable genetic algorithmINTERNATIONAL JOURNAL OF IMAGING SYSTEMS AND TECHNOLOGY, Issue 3 2009Mohamad M. Awad Abstract Image segmentation is an important task in image processing and analysis. Many segmentation methods have been used to segment satellite images. The success of each method depends on the characteristics of the acquired image such as resolution limitations and on the percentage of imperfections in the process of image acquisition due to noise. Many of these methods require a priori knowledge which is difficult to obtain. Some of them are parametric statistical methods that use many parameters which are dependent on image property. In this article, a new unsupervised nonparametric method is developed to segment satellite images into homogeneous regions without any a priori knowledge. The new method is called hybrid variable genetic algorithm (HVGA). The variability is found in the variable number of cluster centers and in the changeable mutation rate. In addition, this new method uses different heuristic processes to increase the efficiency of genetic algorithm in avoiding local optimal solutions. Experiments performed on two different satellite images (Landsat and Spot) proved the high accuracy and efficiency of HVGA compared with another two unsupervised and nonparametric segmentation methods genetic algorithm (GA) and self-organizing map (SOM). The verification of the results included stability and accuracy measurements using an evaluation method implemented from the functional model (FM) and field surveys. © 2009 Wiley Periodicals, Inc. Int J Imaging Syst Technol, 19, 199,207, 2009 [source] Fixed-order H, control design via a partially augmented Lagrangian methodINTERNATIONAL JOURNAL OF ROBUST AND NONLINEAR CONTROL, Issue 12 2003Pierre Apkarian Abstract In this paper we develop an augmented Lagrangian method to determine local optimal solutions of the reduced- and fixed-order H, synthesis problems. We cast these synthesis problems as optimization programs with a linear cost subject to linear matrix inequality (LMI) constraints along with nonlinear equality constraints representing a matrix inversion condition. The special feature of our algorithm is that only equality constraints are included in the augmented Lagrangian, while LMI constraints are kept explicitly in order to exploit currently available semi definite programming (SDP) codes. The step computation in the tangent problem is based on a Gauss,Newton model, and a specific line search and a first-order Lagrange multiplier update rule are used to enhance efficiency. A number of computational results are reported and underline the strong practical performance of the algorithm. Copyright © 2003 John Wiley & Sons, Ltd. [source] Robust Isolation Of Sensor FailuresASIAN JOURNAL OF CONTROL, Issue 1 2003R. Xu ABSTRACT Sensor self-validity check is a critical step in system control and fault diagnostics. In this paper, a robust approach to isolate sensor failures is proposed. First, a residual model for a given system is built off-line and directly based on input-output measurement data. The residual model outputs are called "primary residuals" and are zero when there is no fault. Most conventional approaches to residual model generation are indirect, as they first require the determination of state-space or other models using standard system identification algorithms. Second, a new max-min design of structured residuals, which can maximize the sensitivity of structured residuals with respect to sensor failures, is proposed. Based on the structured residuals, one can then isolate the sensor failures. This design can also be done in an off-line manner. It is an optimization procedure that avoids local optimal solutions. Simulation and experimental results demonstrated the effectiveness of the proposed method. [source] | ||||||||||