No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
The research and advances on some focus questions in computation intelligence
Abstract
A systematic analysis is made on some focus questions in computation intelligence, emphasis is laid on such a few of problems as the mapping and representation, generalization, topologic structure learning, global optimum approximation of multilayer feedforward networks,the genetic algorithms and the simulated annealing algorithms. By means of comparing the simulation process or mathematical proof process with the actual biology process, physics process or mathematical approximation process, the causes of some focus questions,about which a widespread discuss has being made, are discovered for the first time. On one hand ,some interesting questions are presented, on the other hand, some original ideas and methods to solve these questions are also suggested for reference and discussion.
The structural failure is fuzzy as well as random. In this paper, a finite element method and a simulated annealing-genetic algorithm are proposed to determine the structural fuzzy reliability. A modified joint probability density function (MJPDF) is introduced for each failure level such that the most probable point of failure (the design point) is calculated by finding a maximum to this function in the corresponding failure field. The finite element method is used to calculate the load term. Conventional optimization approaches generally have much difficulty in dealing with MJPDF. In this paper the simulated annealing algorithm is incorporated in the generic algorithm to improve the convergence performance of the genetic algorithm. The proposed approach can search for the global optimum more efficiently and effectively, not requiring calculation of numerical derivatives. A fuzzy reliability assessment of a cantilever beam illustrates the application of the method.
According to the demand of practical bidding procedure in the construction field, the paper adapts the K-L information distance to the learning error function of the neural network and improves the traditional BP algorithm. The bidding system's learning ability and generalization ability to various learning algorithms are compared. Simulations illustrate that the improved BP algorithm makes the whole bidding system have better generalization ability and guarantees its validity in the practical application.
The goal of this paper is to study the factors that affect the generalization capability and real-time learning for neural network. First, this paper investigates the effect of initial weight ranges, learning rate, and regularization co—efficient on generalization capability and learning speed. Based on this, this paper proposes a hybrid method that simultaneously considers these three factors, and dynamically tunes the learning rate and regularization coefficient. Then the paper presents the results of some experimental comparison among these kinds of methods in several different problems. Finally, it draws conclusions and makes plan for future work.
ResearchGate has not been able to resolve any references for this publication.