Catálogo de publicaciones - libros

Hopfield introduced the neural network for linear programming with linear constraints. In this paper, Hopfield neural network has been generalized to solve the optimization problems including nonlinear constraints. The proposed neural network can solve a nonlinear cost function with nonlinear constraints. Also, methods have been discussed to reconcile optimization problems with neural networks and implementation of the circuits. Simulation results show that the computational energy function converges to stable point by decreasing the cost function as the time passes.

The application of sinusoidal periodic search signals into the general extremum seeking algorithm(ESA) results in the “chatter” problem of the output and the switching of the control law and incapability of escaping from the local minima. A novel chaotic annealing recurrent neural network (CARNN) is proposed for ESA to solve those problems in the general ESA and improve the capability of global searching. The paper converts ESA into seeking the global extreme point where the slope of Cost Function is zero, and applies a CARNN to finding the global point and stabilizing the plant at that point. ESA combined with CARNN doesn’t make use of search signals such as sinusoidal periodic signals, which solves those problems in previous ESA and improves the dynamic performance of the ESA system greatly. During the process of optimization, chaotic annealing is realized by decaying the amplitude of the chaos noise and the probability of accepting continuously. The process of optimization was divided into two phases: the coarse search based on chaos and the elaborate search based on RNN. At last, CARNN will stabilize the system to the global extreme point. At the same time, it can be simplified by the proposed method to analyze the stability of ESA. The simulation results of a simplified UAV tight formation flight model and a typical testing function proved the advantages mentioned above.

A wavelet function was introduced into the activation function of the transiently chaotic neural network in order to solve combinational optimization problems more efficiently. The dynamic behaviors of chaotic signal neural units were analyzed and the time evolution figures of the maximal Lyapunov exponents and chaotic dynamic behavior were given. The improved transiently chaotic neural network has the ability to stay in chaotic states longer because the wavelet function is non-monotonous and is a kind of basic function. The simulation results prove that the improved transiently chaotic neural network is superior to the original in solving 10-city traveling salesman problem (TSP).

Based on our previously proposed Quantum-behaved Particle Swarm Optimization (QPSO), this paper discusses the applicability of QPSO to integer programming. QPSO is a global convergent search method, while the original Particle Swarm (PSO) cannot be guaranteed to find out the optima solution of the problem at hand. The application of QPSO to integer programming is the first attempt of the new algorithm to discrete optimization problem. After introduction of PSO and detailed description of QPSO, we propose a method of using QPSO to solve integer programming. Some benchmark problems are employed to test QPSO as well as PSO for performance comparison. The experiment results show the superiority of QPSO to PSO on the problems.

Particle swarm optimization is widely applied for training neural network. Since in many applications the number of weights of NN is huge, when PSO algorithms are applied for NN training, the dimension of search space is so large that PSOs always converge prematurely. In this paper an improved stochastic PSO (SPSO) is presented, to which a random velocity is added to improve particles’ exploration ability. Since SPSO explores much thoroughly to collect information of solution space, it is able to find the global best solution with high opportunity. Hence SPSO is suitable for optimization about high dimension problems, especially for NN training.

Training neural networks is a complex task of great importance in problems of supervised learning. The Particle Swarm Optimization (PSO) consists of a stochastic global search originated from the attempt to graphically simulate the social behavior of a flock of birds looking for resources. In this work we analyze the use of the PSO algorithm and two variants with a local search operator for neural network training and investigate the influence of the stop criteria in generalization control for swarm optimizers. For evaluating these algorithms we apply them to benchmark classification problems of the medical field. The results showed that the hybrid GCPSO with local search operator had the best results among the particle swarm optimizers in two of the three tested problems.

Clonal selection theory describes selection, proliferation, and mutation process of immune cells during immune response. In this Artificial Immune System (AIS), We select not only the highest affinity antibody, but also other antibodies which have higher affinity than that of current memory cell during affinity mutation process. Simulation results for pattern recognition show that the improved model has stronger noise immunity ability than other models.

A pattern recognition approach, based on shape feature extraction, is proposed to infer genetic networks from time course microarray data. The proposed algorithm learns patterns from known genetic interactions, such as RT-PCR confirmed gene pairs, and tunes the parameters using particle swarm optimization algorithm. This work also incorporates a score function to separate significant predictions from non-significant ones. The prediction accuracy of the proposed method applied to data sets in Spellman (1998) is as high as 91%, and true-positive rate and false-negative rate are about 61% and 1%, respectively. Therefore, the proposed algorithm may be useful for inferring genetic interactions.

In high bit rate optical fiber communication systems, Polarization mode dispersion (PMD) is one of the main factors to signal distortion and needs to be compensated. Because PMD possesses the time-varying and the statistical properties, to establish an effective control algorithm for adaptive or automatic PMD compensation is a challenging task. Widely used control algorithms are the gradient-based peak search methods, whose main drawbacks are easy being locked into local sub-optima for compensation and no ability to resist noise. In this paper, we introduce particle swarm optimization (PSO), which is an evolutionary approach, into automatic PMD compensation as feedback control algorithm. The experiment results showed that PSO-based control algorithm has unique features of rapid convergence to the global optimum without being trapped in local sub-optima and good robustness to noise in the transmission line that had never been achieved in PMD compensation before.

The nature of construction claims is highly complicated and the cost involved is high. It will be advantageous if the parties to a dispute may know with some certainty how the case would be resolved if it were taken to court. The recent advancements in artificial neural networks may render a cost-effective technique to help to predict the outcome of construction claims, on the basis of characteristics of cases and the corresponding past court decisions. In this paper, a split-step particle swarm optimization (PSO) model is applied to train perceptrons in order to predict the outcome of construction claims in Hong Kong. It combines the advantages of global search capability of PSO algorithm in the first step and the local convergence of back-propagation algorithm in the second step. It is shown that, through a real application case, its performance is much better than the benchmark backward propagation algorithm and the conventional PSO algorithm.