High-order neural network structure selection for function approximation applications using genetic algorithms

Neural network literature for function approximation is by now sufficiently rich. In its complete form, the problem entails both parametric (i.e., weights determination) and structural learning (i.e., structure selection). The majority of works deal with parametric uncertainty assuming knowledge of the appropriate neural structure. In this paper we present an algorithmic approach to determine the structure of High Order Neural Networks (HONNs), to solve function approximation problems. The method is based on a Genetic Algorithm (GA) and is equipped with a stable update law to guarantee parametric learning. Simulation results on an illustrative example highlight the performance and give some insight of the proposed approach.     

GA BP算法在混凝土结构故障诊断中的应用研究*

混凝土结构故障诊断问题包含极其丰富的内容,实际工程中又不断地提出各种新的要求,致使其利用传统的方法难以解决.应用神经网络(NN)与遗传算法(GA)来解决混凝土研究中的难点问题,已经成为当前混凝土结构故障诊断研究领域的一个热门课题.GA-BP网络诊断模型为解决大规模、复杂、并行的系统问题提供了广阔的前景.     

Evolutionary Radial Basis Functions for Credit Assessment

Credit analysts generally assess the risk of credit applications based on their previous experience. They frequently employ quantitative methods to this end. Among the methods used, Artificial Neural Networks have been particularly successful and have been incorporated into several computational tools. However, the design of efficient Artificial Neural Networks is largely affected by the definition of adequate values for their free parameters. This article discusses a new approach to the design of a particular Artificial Neural Networks model, RBF networks, through Genetic Algorithms. It presents an overall view of the problems involved and the different approaches employed to optimize Artificial Neural Networks genetically. For such, several methods proposed in the literature for optimizing RBF networks using Genetic Algorithms are discussed. Finally, the model proposed by the authors is described and experimental results using this model for a credit risk assessment problem are presented.     

遗传前馈神经网络在函数逼近中的应用

人工神经网络具有高计算能力、泛化能力和非线性映射等特点,被成功应用于众多领域,但缺乏用于确定其网络拓扑结构、激活函数和训练方法的规则。该文提出利用遗传算法优化前馈神经网络的方法,将网络结构、激活函数和训练方法等编码作为个体,发现最优或次优解,针对特定问题设计较理想的前馈神经网络。介绍遗传算法的具体步骤,对非线性函数逼近进行实验,结果表明优化后前馈神经网络的性能优于由经验确定的前馈神经网络,验证了本文方法的有效性。     

单亲遗传算法及其全局收敛性分析

序号编码的遗传算法(GA)不能在两条染色体的任意位置进行交叉,必须使用 PMX,CX和OX等特殊的交叉算子,而这些交叉算子实施起来都很麻烦.针对序号编码GA 的上述不足,提出一种单亲遗传算法(PGA).PGA采用序号编码,不使用交叉算子,而代之以 隐含序号编码GA交叉算子功能的基因换位等遗传算子,简化了遗传操作,并且不要求初始 群体具有多样性,也不存在"早熟收敛"问题.仿真结果验证了这种算法的有效性.     

基于EPNET模型和多期预测的时间序列预测

文中主要研究了基于EPNET(Evolutionary Programming Net)的时间序列预测问题。EPNET是一种进化人工神经网络模型，它能够同时代化网络权值和网络结构。该模型没有采用遗传算法中的交叉算子，而是采用了五个变异算子来获得比较理想的进化效果。在此基础上，提出了基于该模型的时间序列预测算法，介绍了该算法实现时的有关问题。     

Hybrid neuro-genetic based method for solving ill-posed inverse problem occurring in synthesis of electromagnetic fields

Fredholm Integral Equation of the First Kind (FIEFK) is an example of ill-posed problems. Solving this type of equation using conventional methods of discretization often leads to an ill-conditioned system of linear equations. This paper deals with the numerical solution for the FIEFK occurring in the synthesis of the electromagnetic fields. To tackle this problem, we propose a hybrid method based on Genetic Algorithms (GAs) and Artificial Neural Networks. The method consists of two major steps. The first step is to find an initial solution by utilizing a GA, and the second is to refine the solution using a regularized neural network. Experimental results prove the efficiency of our proposed method in comparison with a previous work.     

改进的径向基函数网络的研究及应用

针对径向基函数网络和传统遗传算法的一些不足,提出引入一种自适应机制的浮点数编码的遗传算法,并将其与梯度下降法混合交互运算,作为径向基函数网络的学习算法,形成了基于改进遗传算法的径向基函数网络,它克服了径向基函数网络的学习算法上的缺陷。采用改进的遗传算法,无需计算梯度等,限制很少,还可用模型的预测性能作为优化目标。同时,也解决了单独利用遗传算法往往只能在短时间内寻找到接近全局最优解的近似解这一问题。最后将该算法应用到某地区电力负荷预测取得理想效果。     

The multi-criteria minimum spanning tree problem based genetic algorithm

Minimum spanning tree (MST) problem is of high importance in network optimization and can be solved efficiently. The multi-criteria MST (mc-MST) is a more realistic representation of the practical problems in the real world, but it is difficult for traditional optimization technique to deal with. In this paper, a non-generational genetic algorithm (GA) for mc-MST is proposed. To keep the population diversity, this paper designs an efficient crossover operator by using dislocation a crossover technique and builds a niche evolution procedure, where a better offspring does not replace the whole or most individuals but replaces the worse ones of the current population. To evaluate the non-generational GA, the solution sets generated by it are compared with solution sets from an improved algorithm for enumerating all Pareto optimal spanning trees. The improved enumeration algorithm is proved to find all Pareto optimal solutions and experimental results show that the non-generational GA is efficient.     

面向对象的遗传算法及其在神经网络辅助设计中的应用

在现有的遗传算法的基础上,采用面向对象技术设计了面向对象的遗传算法,建立了遗传算法的类层次。这种方法改变了在传统的遗传算法中各个函数之间只有参数的传递,而没有代码的继承性的状况从概念上提高了软件的可重用性。该方法在人工神经网络的辅助设计问题中的应用表明,这一算法由于采用面向对象的分析与设计方法,从而具有比传统的遗传算法更好的通用性,用户可以更方便地设计和实现自己的编码方案和遗传算子,大大提高了软件的可重用性。     

A neuro-fuzzy model for a new hybrid integrated Process Planning and Scheduling system

In customized mass production, isolation of Process Planning (PP) and Scheduling stages has a critical effect on the efficiency of production. In this study, to overcome this isolation problem, we propose an integrated system that does PP and Scheduling in parallel and responds to fluctuations in job floor on time. One common problem observed in integration models is the increase in computational time in conjunction with the increase of problem size. Therefore in this study, we use a hybrid heuristic model combining both Genetic Algorithm (GA) and Fuzzy Neural Network (FNN). To improve GA performance and increase the efficiency of searching, we use a clustered chromosome structure and test the performance of GA with respect to different scenarios. Data provided by GA is used in constructing an FNN model that instantly provides new schedules as new constraints emerge in the production environment. Introduction of fuzzy membership functions in Artificial Neural Network (ANN) model allows us to generate fuzzy rules for production environment.     

Structural topology design optimization using Genetic Algorithms with a bit-array representation

In this paper, a bit-array representation method for structural topology optimization using the Genetic Algorithm (GA) is implemented. The importance of structural connectivity in a design is further emphasized by considering the total number of connected objects of each individual explicitly in an equality constraint function. To evaluate the constrained objective function, Deb’s constraint handling approach is further developed to ensure that feasible individuals are always better than infeasible ones in the population to improve the efficiency of the GA. A violation penalty method is proposed to drive the GA search towards the topologies with higher structural performance, less unusable material and fewer separate objects in the design domain. An identical initialization method is also proposed to improve the GA performance in dealing with problems with long narrow design domains. Numerical results of structural topology optimization problems of minimum weight and minimum compliance designs show the success of this bit-array representation method and suggest that the GA performance can be significantly improved by handling the design connectivity properly.     

A genetic algorithm for finding a path subject to two constraints

Multi-constrained routing (MCR) aims to find the feasible path in the network that satisfies multiple independent constraints, it is usually used for routing multimedia traffic with quality-of-service (QoS) guarantees. It is well known that MCR is NP-complete. Heuristic and approximate algorithms for MCR are not effective in dynamic network environment for real-time applications when the state information of the network is out of date. This paper presents a genetic algorithm to solve the MCR problem subject to transmission delay and transmission success ratio. Three key design problems are investigated for this new algorithm, i.e., how to encode the problem in genetic representation, how to avoid the illegal chromosomes in the process of population initialization and genetic operation, and how to design effective genetic operator. We propose the gene structure (GS) to deal with the first problem, and the gene structure algorithm (GSA) to generate the GS. Based on the GS, we provide the heuristic chromosome initialization and mutation operator to solve the last two problems. Computer simulations show that the proposed GA exhibits much faster computation speed so as to satisfy the real-time requirement, and much higher rate of convergence than other algorithms. The results are relatively independent of problem types (network scales and topologies). Furthermore, simulation results show that the proposed GA is effective and efficient in dynamic network environment.     

一种改进选择算子的遗传算法

遗传算法（Genetic Algorithm,GA）是一种模拟生物进化的智能算法,被广泛应用于求解各类问题。简单遗传算法（Simple GA）仅靠变异产生新的数值,常常存在搜索精确度不高的问题。针对这个问题,对SGA的选择算子进行改进,即把相似个体分在同一组中,以组为单位进行选择,并通过该组个体的特点进行高斯搜索生成新的群体。这样使得GA在搜索过程中不仅可以很好地保持个体的多样性,并且可以提高解的精确度。通过对11个函数（单峰和多峰）的仿真实验,证明了采用新的选择算子后,GA在求解问题的精确度上有了很大地改善。     

A hierarchical parallel genetic approach for the graph coloring problem

Graph Coloring Problems (GCPs) are constraint optimization problems with various applications including time tabling and frequency allocation. The GCP consists in finding the minimum number of colors for coloring the graph vertices such that adjacent vertices have distinct colors. We propose a hierarchical approach based on Parallel Genetic Algorithms (PGAs) to solve the GCP. We call this new approach Hierarchical PGAs (HPGAs). In addition, we have developed a new operator designed to improve PGAs when solving constraint optimization problems in general and GCPs in particular. We call this new operator Genetic Modification (GM). Using the properties of variables and their relations, GM generates good individuals at each iteration and inserts them into the PGA population in the hope of reaching the optimal solution sooner. In the case of the GCP, the GM operator is based on a novel Variable Ordering Algorithm (VOA) that we propose. Together with the new crossover and the estimator of the initial solution we have developed, GM allows our solving approach to converge towards the optimal solution sooner than the well known methods for solving the GCP, even for hard instances. This was indeed clearly demonstrated by the experiments we conducted on the GCP instances taken from the well known DIMACS website.     

Automatic diagnosis with genetic algorithms and case-based reasoning

This article describes the application of Machine Learning (ML) techniques to a real world problem: the Automatic Diagnosis (classification) of Mammary Biopsy Images. The techniques applied are Genetic Algorithms (GA) and Case-Based Reasoning (CBR). The article compares our results with previous results obtained using Neural Networks (NN). The main goals are: to efficiently solve classification problems of such a type and to compare different alternatives for Machine Learning. The article also introduces the systems we developed for solving this kind of classification problems: Genetic Based Classifier System (GeB-CS) for a GA approach, and Case-Based Classifier System (CaB-CS) for a CBR approach.     

An effective evolutionary algorithm for the practical capacitated vehicle routing problems

The evolutionary algorithms are extensively adopted to resolve complex optimization problem. Genetic algorithm (GA), an evolutionary algorithm, has been proved capable of solving vehicle routing problems (VRPs). However, the resolution effectiveness of GA decreases with the increase of nodes within VRPs. Normally, a hybrid GA outperforms pure GA. This study attempts to solve a capacitated vehicle routing problem (CVRP) by applying a novel hybrid genetic algorithm (HGA) that is practical for use by manufacturers. The proposed HGA involves three stages. First, a diverse and well-structured initial chromosome population was constructed. Second, response surface methodology (RSM) experiments were conducted to optimize the crossover and mutation probabilities in performing GA. Finally, a combined heuristics containing improved insertion algorithm and random insertion mutation operator was established to stir over gene permutations and enhance the exploration capability of GA diversely. Furthermore, an elitism conservation strategy was implemented that replace inferior chromosomes with superior ones. As the proposed HGA is primarily used to solve practical problems, benchmark problems involving fewer than 100 nodes from an Internet website were utilized to confirm the feasibility of the proposed HGA. Two real cases one for locally active distribution and another for arms part transportation at a combined maintenance facility, both involving the Taiwanese armed forces are used to detail the analytical process and demonstrate the practicability of the proposed HGA for optimizing the CVRP.     

Genetic optimized artificial immune system in spam detection: a review and a model

Spam is a serious universal problem which causes problems for almost all computer users. This issue affects not only normal users of the internet, but also causes a big problem for companies and organizations since it costs a huge amount of money in lost productivity, wasting users’ time and network bandwidth. Many studies on spam indicate that spam cost organizations billions of dollars yearly. This work presents a machine learning method inspired by the human immune system called Artificial Immune System (AIS) which is a new emerging method that still needs further exploration. Core modifications were applied on the standard AIS with the aid of the Genetic Algorithm. Also an Artificial Neural Network for spam detection is applied with a new manner. SpamAssassin corpus is used in all our simulations.     

Derivative and GA-based methods in metamodeling of back-propagation neural networks for constrained approximate optimization

Artificial neural networks (ANN) have been extensively used as global approximation tools in the context of approximate optimization. ANN traditionally minimizes the absolute difference between target outputs and approximate outputs thereby resulting in approximate optimal solutions being sometimes actually infeasible when it is used as a metamodel for inequality constraint functions. The paper explores the development of the efficient back-propagation neural network (BPN)-based metamodel that ensures the constraint feasibility of approximate optimal solution. The BPN architecture is optimized via two approaches of both derivative-based method and genetic algorithm (GA) to determine interconnection weights between layers in the network. The verification of the proposed approach is examined by adopting a standard ten-bar truss problem. Finally, a GA-based approximate optimization of suspension with an optical flying head is conducted to enhance the shock resistance capability in addition to dynamic characteristics.     

基于混合遗传神经网络的百米跑成绩预测方法

在遗传算法(Genetic ALgorithm)与BP(Back Propagation)网络结构模型相结合的基础上，设计了用遗传算法训练神经网络权重的新方法，并把这种方法用于运动员百米跑成绩预测。与BP算法和LM(Levenberg Marquardt)算法相比，基于混合遗传算法的神经网络不仅有较快的学习速度和较好的学习精度，而且网络的泛化能力(Generalization Ability)得到了很大提高。