Support Vector Regression And Ant Colony Optimization For Grid Resources Prediction

In grid resource prediction, Grid resource prediction is the basis for resource management for efficient task scheduling, dynamic load balancing is very important, but also will affect the performance of the tasks running on the grid. Many relevant research models have been developed and have generated accurate prediction in practice. These prediction models that provide future resources information generally apply the time series prediction models which mostly use statistical and artificial intelligent methods.Many domestic and overseas researchers used traditional statistical model for grid resource prediction. They used AR, MA, ARMA, ARIMA, ARFIMA model for grid resource prediction, and achieved good results. However, traditional statistical model for nonlinear time series prediction is not satisfactory, while the grid resources over time is often non-linear, a number of artificial intelligence methods are applied to resource prediction successfully. Artificial neural network as the most commonly used artificial intelligence method has good nonlinear approximation ability and adaptive self-learning functions in grid resource prediction has been more successful applications. However, the existence of artificial neural networks need to identify the network structure, training samples required large, than to learn and easy to fall into local minimum and so on.Support vector machine (SVM) that overcomes the shortcomings of artificial neural networks is widely used to predict the problems. SVM method first was used to solve classification problems. With the introduction ofεloss function, support vector regression (SVR) algorithm was introduced. This approach used structural risk minimization principle to replace the traditional machine learning methods in the empirical risk minimization principle, and has a good generalization ability of machine learning in small samples, which showed excellent performance. Support vector regression has intuitive geometric interpretation: the existence of a width of 2? (? is a constant) of a pipeline, pipeline center is seeking to be the optimal hyperplane. The pipeline attempts to include as many data points to this tract, including to the point that the pipeline was accurately predicted. Classic regression method is to find a curve, so that the line through the points as much as possible. This leads to the problem that the curve found is too complex to be accurately indicate the overall trend of data points, especially encountered when noise and outliers are too easy to learn. But if a pipeline through the points as much as possible to the data points would capture the overall trend, not easy to excessive learning, from the geometric meaning of this can be seen on the advantages of support vector regression. According to the above analysis, we could get the reasons why SVM outperforms ANN:First, compared with BP network, SVR model has a better non-linear mapping ability, can easily grasp the data model.Secondly, SVR model follows the principle of structural risk minimization rather than empirical risk minimization principle, and effectively improves the generalization performance of the model.Thirdly, the solution of SVM is sparse. This character is what ANN does not hold. It is a reason why SVM works efficiently.This prediction model is based on support vector regression algorithm. In practice, SVR's performance depends on its free parameters. In the absence of theoretical guidance to do the traditional SVR parameters are selected through the test method, a satisfactory solution of artificial selection. This approach depends on the person's subjective experience, and it's quite time-consuming. Ant Colony Optimization (ACO) algorithm is a typical evolutionary computing stochastic optimization method. In this paper, ant colony optimization algorithm for global search capability, the SVR parameters automatically determined by artificial selection becomes, reducing the blindness of parameter selection and to improve the accuracy of the parameters. Experiments showed that the proposed ant colony optimization algorithm to optimize the parameters of support vector regression prediction model in grid resource prediction got a more satisfactory prediction.Grid resource prediction plays a very important role in grid scheduling. This article is the original by improving the ant colony algorithm to make up for the big error caused by the original algorithm, easy to premature, and then improved ant colony optimization algorithm to optimize the support vector regression model of key parameters, makes the model can be on the standard data the optimal choice, be minimal error, improve forecasting accuracy, provide the basis for resource scheduling. Experiments showed that the proposed ACO-SVR model could have higher precision of grid resource prediction, than the traditional test method. It also had higher efficiency and prediction accuracy, indicating that the ant colony optimization algorithm optimizing SVR parameters was success.Experimental results show the ACO-SVR model in grid resource prediction enormous potential. In the next work, we will try some other heuristic optimization algorithm to determine the parameters of SVR and look forward to achieve better results.This study results will be helpful to a grid infrastructure and play an important role in future research work. The future work is as follow:1. To further improve our work.This work used ACO-SVR model in the general resources of the grid nodes to predict the lack of an overall prediction grid resources into account; due to the diversity of Grid resources themselves so that we must address a more extensive grid resources to provide more excellent forecasting model, so that in the field of grid computing applications to play a bigger role.2. Consider other models for grid resource forecasting.Parallel genetic algorithm and parallel particle swarm optimization algorithm can be considered to improve the efficiency of parameter optimization. The key problems are: How to seasonally divide the search space, so that different machines in different search space for optimization, optimization of individual machines and then take the best results. This will make the best use of distributed grid characteristics, effectively improve forecasting performance.3. Application of other evolutionary algorithms to optimize the model parameters to predict.This forecast system is only achieved by using the ant colony optimization algorithm. At the same time there are many other optimization algorithms, such as genetic algorithms, particle swarm algorithms. These algorithms'framework now also has been extensively studied, but there are a lot of studies have shown that these algorithms in a number of occasions, the performance of ant colony algorithm is superior to the ordinary, try these new algorithms to optimize the application of support vector machines will also significant.