The Research Of Some Optimization Problems Based On Computational Intelligence

Optimization technique is based on mathematics and used to find the optimumsolutions of many engineering problems. Since 1980's, the industry control has beendeveloping in the direction of large-scale, sequential, compositive and intricate process.Owing to the feature of complexity, constraint, nonlinearity, multi-local minimum anddifficulty to model, optimality calculation of all kinds of process control is an urgentproblem that needs to be solved. So it has become a primary research direction in the relatedsubject and field to research intelligent algorithms for solving large-scale parallel problems.The research of intelligent optimization algorithms is currently a focus in the advancedfield. These algorithms include artificial neural network (ANN), genetic algorithm (GA),simulated annealing (SA), particle swarm optimization (PSO), ant colony optimization(ACO), artificial immune system (AIS), etc., which are developed by simulating orrevealing some phenomenon and process of nature. They provide innovative thought andmeans to solve many complicated problems. These algorithms have attracted broad interestsof many researchers in the world for their particular advantage and mechanism and havebeen applied to many fields successfully. At the same time, traditional optimizationmethods are difficulty to solve complicated large-scale engineering problems, because theyare generally difficulty to satisfy the requirement of the computation speed, convergenceand sensitivity of initial values. Whereas the optimization algorithms based on computationintelligence are easy to realize and calculate, which involve only the basic mathematicoperations, especially, most of computational intelligence methods possess latent paralleland distribution mechanism, which provide comprehensive technique guarantee to deal witha mass of data stored in database. Therefore, the study of intelligence optimizationalgorithms has important theoretical and practical significance. This paper researches someoptimization problems based on the approaches of computation intelligence. The majorcontents could be summarized as follows:1. The design purpose of a control system is to influence the behavior of dynamicsystems to achieve some predeterminated objectives. The control is usually requested toimplement on the premise that the accurate object and environmental knowledge cannot beobtained in advance. So it is also desired to find suitable methods to solve the problems ofuncertain and highly complicated dynamic system identification. However, in the field ofautomatic control, the methods of system identification and parameter amendment based onthe linear analysis can only be applied to linear systems in most cases. It is difficult toextend the linear-based methods to the complicated non-linear systems. An ultrasonic motor(USM) is a typical non-linear dynamic system, which has some excellent performances anduseful features. The USM has attracted considerable attention in many practical applications.The simulation and control of the USM are important problems in the applications of theUSM. According to the conventional control theory, an accurate mathematical model shouldbe set up. But the USM has strongly nonlinear speed characteristics that vary with thedriving conditions and its operational characteristics depend on many factors. Therefore, itis difficult to perform effective identification and control to the USM using traditionalmethods based on mathematical models of systems. In this paper, a learning algorithm fordynamic recurrent Elman neural networks is proposed based on an improved adaptive GA.Two dynamic identification algorithms for nonlinear systems are constructed successivelybased on the proposed algorithm to perform the speed identification for the USM, and thecomplex theoretical analysis of the operational mechanism and the exact mathematicaldescription of the USM are avoided. Numerical results show that the proposed algorithmsnot only realize the fully automatic optimization design for the dynamic recursive neuralnetwork, but also improve the precision of convergence for model identification. But usingGA to train dynamic recurrent networks needs a tremendous computational effort, and itdoesnot address such dynamic problems with a high requirement of real-time quality. So alearning algorithm for dynamic recurrent Elman neural networks based on a dissimilationPSO is proposed. The algorithm can approximate the optimum solution very fast.Successively, a novel dynamic identifier is constructed to perform speed identification andalso a controller is designed to perform speed control for the USM. Numerical results showthat the designed identifier and controller based on the proposed algorithm can both achievehigher convergence precision and speed. The identifier can approximate the nonlinearinput-output mapping of the USM quite well, and the good control effectiveness of thecontroller is verified using different kinds of speeds of constant, step, and sinusoidal types.Besides, the preliminary examination on the randomly perturbation also shows the fairlyrobust characteristics of the two models.2. An artificial neural network (ANN) is defined as a structure composed of a numberof interconnected units or artificial neurons, and it is a high non-linear system. The ANNapproach has a good potential for identification and control applications because it canapproximate the nonlinear input-output mapping of a dynamic system. In fact, in recentyears, there has been a growing interest in the application of neural networks to dynamicsystem identification and control. The "depth" and "resolution ratio" are the primary indexused to evaluate the memory characteristic of a neural network. The memory of a delay unitis low depth and high resolution ratio, whereas the memory of a recurrent neural network isgenerally high depth and low resolution ration. So the memory performance of most neuralnetworks has limitation in some degree. This paper proposes a time-delay recurrent neuralnetwork with high depth and resolution ratio by introducing the time-delay and recurrentmechanism. The learning algorithm is derived according to gradient descent method. Inorder to guarantee the convergence of the models, the adaptive learning rates of neuralnetworks are derived using discrete-type Lyapunov stability analysis. Numeral experimentsfor identifying nonlinear time-varying systems and controlling bilinear DGP systems as wellas inverted pendulum systems show that the TDRNN has good effectiveness in theidentification and control for dynamic systems.3. The optimization problems are divided two classes of continuous variable anddiscrete variable, and the optimization problem of discrete variable is namely combinatorialoptimization problem. The Traveling Salesman Problem (TSP) is a well-knownnon-polynomial time or NP-hard combinatorial optimization problem that has manypotential applications. Early efforts in combinatorial optimization in the field ofevolutionary computation focused on the TSP, which is a generalization and simplified formof various complicated problems that have appeared in many fields. Hence, a successfulsolution for the TSP has not only theoretical significance in computational theory but alsoimportant practical value. In this paper, several expanded TSPs are discussed, and somenovel improved genetic operations are presented. Several combinations of geneticoperations are examined and the functions of these operations at different evolutionarystages are analyzed by numerical experiments. The essentiality of the ordering of the genesection, the significance of the evolutionary inversion operation and the importance of theselection model are discussed. Some results and conclusions are obtained and presented,which provide useful information for the implementation of the genetic operations forsolving the traveling salesman problem.4. The acceleration in the accumulation of biological sequences is more and more great,which makes it challenging to search for sequence similarity by comparing many relatedsequences simultaneously. Multiple sequence alignment (MSA) is one of the most importanttechniques in the sequence comparison. Algorithms for multiple sequence alignment arecommonly used to find conserved regions in biomolecular sequences, to construct familyand superfamily representations of sequences, and to reveal evolutionary histories of species.Multiple sequence alignment is NP-hard problem. A very general form of probabilisticmodel for sequences of symbols, called a hidden Markov model (HMM), has been appliedto MSA. The critical and difficult problem for using HMM approach is how to set up asteady and reliable model by a finite training set, moreover, there is no known deterministicalgorithm that can guarantee to find the optimally trained HMM with reasonable runningtime. In this paper, an immune particle swarm optimization (IPSO) is proposed, which isbased on the models of the vaccination and the receptor editing in immune systems. Theproposed algorithm is used to train the HMM, further, an integration algorithm based on theHMM and IPSO for the MSA is constructed. The proposed algorithm possesses therandomicity of stochastic optimization algorithms, which can be used to solve theoptimization problem for non-linear systems;moreover, the proposed algorithm possessesthe adaptive ability that enables the algorithm to solve machine learning problems. Theapproach is tested on a set of standard instances taken from the Benchmark alignmentdatabase, BAliBASE. Numerical simulated results show that the proposed algorithm notonly improves the alignment abilities, but also reduces the time cost, which provide aneffective approach for the MSA.5. The job shop scheduling problem (JSSP) is a very important practical problem inboth fields of production management and combinatorial optimization. Efficient methods forsolving the JSSP have significant effects on profitability and product quality. The JSSP hasdrawn the attention of researchers for the last three decades, but because schedulingproblems vary widely according to specific production tasks and most are strongly NP-hardproblems such that some test problems of moderate size are still unsolved. There remainsmuch room for the improvement in current techniques and the exploitation in new efficientmethods. In this paper, four novel scheduling algorithms are proposed, which arerespectively based on particle swarm optimization (PSO), artificial immune system (AIS),the hybrid algorithm of simulated annealing (SA) and ant colony optimization (ACO), andthe hybrid algorithm of PSO and AIS. In the particle swarm system, a novel concept for thedistance and velocity of particles is expanded to pave the way for the JSSP. In the artificialimmune system, the models of vaccination and receptor editing are designed to improve theimmune performance. In the ant system, the definition of bidirectional constraint link graphis proposed to address the algorithm of meet max-min ant system (MMMAS). Furthermore,some assistant handling technologies are also proposed, such as the trace of step memory,initial schedule generation algorithm, δ inching tuning, and so on. These algorithms aretested on a set of benchmark instances with various sizes and levels of hardness. Simulationresults show that the proposed algorithms can obtain high quality solutions with reasonabletime, especially, the hybrid algorithm of the PSO and AIS.