Fast genetic on-line learning algorithm for neural network and its application to temperature control

The Sixth International Symposium on Neural Networks (ISNN 2009)

A way in which a neural network can implement a genetic algorithm as its learning algorithm is shown. This model is called GLANN (genetic learning algorithm for neural networks). The components of GLANN can be shown to be biologically plausible. The algorithm itself can be classified as a reinforcement learning algorithm. The neural network has a fixed architecture and processes binary strings using genetic operators. Learning is stored in the form of newly created patterns, which can then be stored in some kind of associative memory. The benefits of GLANN reside in the proven optimizing capabilities of genetic algorithms, and in its parallel implementation. The shallow two-level architecture translates into system scalability, an issue that has not been successfully resolved in the case of other neural network algorithms. >

HighlightsWe propose five spraying parameters according to the characteristics of pig carcasses in the spray-chilling process.A prediction model for pig carcass weight loss, based on a genetic algorithm back-propagation neural network, is proposed to reveal the relationship between weight loss and spraying parameters.To study the effects of various spraying parameters on weight loss, an automatic spray-chilling device was designed, which can modify up to five spraying parameters.Abstract. Because the weight loss of a pig carcass in the spray-chilling process is easily affected by the spraying frequency and duration, a prediction model for weight loss based on a genetic algorithm (GA) back-propagation (BP) neural network is proposed in this article. With three-way crossbred pig carcasses selected as the test materials, the duration and time interval of high-frequency spraying, the duration and time interval of low-frequency spraying, and the duration of a single spray were selected as inputs to the network model. The weight and threshold of the network were then optimized by the GA. The prediction model for pig carcass weight loss established by the GA BP neural network yielded a correlation coefficient of R = 0.99747 between the network output value of the test samples and the target value. Weight loss prediction by the model is feasible and allows better expression of the nonlinear relationship between weight loss and the main controlling factors. The results can be a reference for chilled meat production. Keywords: BP neural network, Genetic algorithm, Pig carcass, Predictive model, Weight loss

This paper presents the development of several efficient learning algorithms for neural networks (ELEANNE). The ELEANNE 1 and ELEANNE 2 are two recursive least-squares learning algorithms, proposed for training single-layered neural networks with analog output. This paper also proposes a new optimization strategy for training single-layered neural networks, which provides the basis for the development of a variety of efficient learning algorithms. This optimization strategy is the source of the ELEANNE 3, a second-order learning algorithm for training single-layered neural networks with binary output. A simplified version of this algorithm, called ELEANNE 4, is also derived on the basis of some simplifying but reasonable assumptions. The two algorithms developed for single-layered neural networks provide the basis for the derivation of ELEANNE 5 and ELEANNE 6, which are proposed for training multilayered neural networks with binary output. The ELEANNE 7 is an efficient algorithm developed for training multilayered neural networks with either binary or analog output. >

In this paper, Dynamic Back Propagation (DBP) is reinvestigated in the context of indirect inverse control of unknown nonlinear dynamic systems using Artificial Neural Networks (ANNs). The overall control scheme includes two networks. The first is used as an approximating model of the unknown nonlinear dynamic behavior of the plant while the second is the controller. The trajectory error is backpropagated through the neural model using DBP algorithm to obtain the control input error. The controller, which is a recurrent ANN, is then adapted also using DBP algorithm. Simulation results obtained with both Standard Back Propagation (SBP) and DBP algorithms for a fourth order nonlinear system clearly show some advantage of DBP.

In order to improve the efficiency and accuracy of thermal and moisture comfort prediction of underwear, a new prediction model is designed by using principal component analysis method to reduce the dimension of related variables and eliminate the multi-collinearity relationship between variables, and then inputting the converted variables into genetic algorithm (GA) and BP neural network. In order to avoid the problems of slow convergence speed and easy falling into local minimum of Back Propagation (BP) neural network, this paper adopted GA to optimize the weights and thresholds of BP neural network, and utilized MATLAB software to program, and established the prediction models of BP neural network and GA–BP neural network. To verify the superiority of the model, the predicted result of GA–BP, PCA–BP and BP are compared with GA–BP neural network. The results show that PCA could improve the accuracy and adaptability of GA–BP neural network for thermal and moisture comfort prediction. PCA–GA–BP model is obviously superior to GA–BP, PCA–BP, BP, SVM and K-means prediction models, which could accurately predict thermal and moisture comfort of underwear. The model has better accuracy prediction and simpler structure.

Given N data pairs {X i , y i }, i=1,2,...,N, where each X i is an n-dimensional vector of independent variables \((X_i = )\) and y i is a dependent variable, the function approximation problem (FAP) is finding a function that best explains the N pairs of X i and y i . From the universal approximation theorem and inherent approximation capabilities proved by various researches, artificial neural networks (ANNs) are considered as powerful function approximators. There are two main issues on the feedforward neural networks’ performance. One is to determine its structure. The other issue is to specify the weights of a network that minimizes its error. Genetic algorithm (GA) is a global search technique and is useful for complex optimization problems. So, it has been considered to have potential to reinforce the performance of neural networks. Many researchers tried to optimize the weights of networks using genetic algorithms alone or combined with the backpropagation algorithm. Others also tried to find a good topology that is even more difficult and called a “black art.”

Composite beams, plates and shells are widely used in the aerospace industry because of their advantages over the commonly used isotropic structures especially when it comes to weight savings. Buckling analyses of composite structural components must be performed in order to ensure, for instance, that a composite panel designed to be part of a control surface does not buckle thereby compromising its aerodynamic shape. Optimization of composite structures has been performed in this paper using Genetic algorithm. Genetic algorithm (GA) approaches are successfully implemented for the TSP. The buckling load of composite plate, which is obtained by the Artificial Neural Networks, was used as the fitness function in the GA to find its optimized value by arranging the ply stacking sequence.

In this paper, a neuro-genetic approach is proposed for the inverse kinematics problem solution of robotic manipulators. The proposed solution method is based on using neural networks and genetic algorithms in a hybrid system. Neural networks have been used by many researchers in the inverse kinematics solution. Since the neural networks work with an acceptable error, the error at the end of learning has to be minimized for sensitive applications. This study is based on using genetic algorithms to minimize this error. A case study is presented for a 6 degree of freedom robot. In the neural network part, three Elman networks are separately trained and then used in parallel since one Elman network may give better result than the other two ones. These three results are placed in the initial population of the genetic algorithm. The end effector position error is defined as the fitness function and genetic algorithm is implemented. Thus, the error is reduced in micrometer levels. Key words: Elman neural networks, error minimization, six-degree-of-freedom robot, genetic algorithms, robotics.

Energy-efficient virtual machine placement in data centres via an accelerated Genetic Algorithm with improved fitness computation

In order to improve the efficiency and accuracy of the prediction of expressway traffic flow, this paper, based on the characteristics of the data of the expressway traffic flow, focuses on an optimized method of prediction with the application of the neural network with genetic algorithm. Applying genetic algorithm, optimizing BP neural network structure and establishing a new mixed model, this algorithm speed up the slow convergence velocity of traditional BP neural network prediction and increases the possibility to escape local minima. This algorithm based on the optimized genetic neural network predicts the actual data of the expressway traffic flow, the result of which shows that the application of the optimized method of prediction with the genetic neural network algorithm is effective and that it improves the rate and the accuracy of the prediction of the expressway traffic flow.

Fisher’s iris data constitute a hard benchmark for clustering procedures, and attracted much work based on statistical methods and new approaches related to evolutionary algorithms and neural networks. We suggest two genetic algorithms effective for simultaneously determining both the optimal number of groups and the assignment of items to groups. The grouping genetic algorithm proposed by Falkenauer (1998) forms the basis of our method, where the variance ratio criterion and the Marriott’s method provide two fitness functions that both allow for fast computation and include the number of groups explicitly as a parameter. Specialized crossover operators, specific for each of the two fitness functions, are designed to accelerate convergence and minimize the number of iterations. Some simple implementations of our genetic algorithms are presented, that allow to classify correctly as many iris plants as the best alternative procedures proposed for this data set. Therefore genetic algorithms seem to constitute a good alternative choice for handling clustering problems.KeywordsClassificationCrossoverEncodingEvolutionary ComputationFitness FunctionMutation

Mobile robot global path planning in a static environment is an important problem. The paper proposes a method of global path planning based on neural network and genetic algorithm. We constructed the neural network model of environemntal information in the workspace for a robot and used this model to establish the relationship between a collision avoidance path and the output of the model. Then the two-dimensional coding for the path via-points was converted to one-dimensional one and the fitness of both the collision avoidance path and the shortest distance are integrated into a fitness function. The simulation results showed that the proposed method is correct and effective.

Among the emerging technologies nowadays, the genetic algorithm, a powerful optimization technique, is becoming the subject of new craze among neural network researchers. Genetic algorithms (GAs) for training and designing artificial neural networks (ANNs) have proved to be a useful integration. This paper reports an improvement over earlier work on the genetic evolution of neural network weights using the two-parents multipoint restricted crossover (Double-MRX) operator proposed by Srivastava, Shukla and Srivastava (Microelectronics Journal, vol. 29, no. 11, p.921-31, 1998). In this research, a methodology to improve network convergence is presented by introducing a new concept contrary to natural law, i.e. crossover with randomly selected multiple crossover sites restricted to lie within individual weight boundaries, hence termed as Triple-MRN. In GAs, the search strategy relies more on exchange of information between individual building blocks by exploiting crossover operator. The use of Triple-MRX promotes cooperation among individuals, that better exploits the new genotypic information contained in genome variation. This ensures a much more effective search, both in terms of quality of the solution and speed of convergence as shown by the simulation experiments. Fitness function used in the authors' study is 1/MSE (mean square error). The effectiveness of the proposed technique is tested by evaluating the capability of neural network to learn a real-world gas identification problem.

In this article we developed a method for optimizing the structure of a fuzzy artificial neural networks through genetic algorithms. This genetic algorithm is used by optimizing the number of weight connections in a neural network structure, by the evolution of those structures as individuals in a population. It is found that the optimization of the neural network provides higher confidence accuracy of the suggested solution in a case based diagnostic system. The computational cost of the optimized network also improved considerably high.