Parameter Optimization For Support Vector Regression Based On Multiple Intelligent Computation Methods

Parameter Optimization for Support Vector Regression Based on Multiple Intelligent Computation MethodsThe birth of the support vector regression machine model which stand on the foundation of the development of statistical learning theory has attracts much more and more attention in pattern recognition, data analysis, data mining and widespread applications in many fields. Relative to the classical regression analysis algorithm and neural network method on empirical risk minimization theory, support vector regression machine, which has good nature of structural risk minimization feature prevents the disaster of dimensionality and over-fitting and so on. This model is very good at dealing with complex high-dimensional non-linear multiple regression problem. In practice, the choice of support vector regression parameters of the model itself is a complex and difficult problem, the domain of the parameters choice demonstrates the characteristics of complex and meaningless distribution, It's difficult to use conventional methods directly to solve this problem. On the other hand, support vector regression machine parameters have a direct impact on the accuracy and predictive power of regression model. Nowadays the commonly method used in parameter selection is either chose parameters too slow, or choose parameters too ineffective, or too unstable. Therefore, in application, there exists calling for a way of support vector regression machine parameters selection algorithm that is widely recognized as fast, accurate, and stable.Intelligent computing methods, especially evolutionary algorithms, are widely used in complex environments finding the optimal solution of complex problems at acceptable time price. At present, there already are some intelligent methods of calculating support vector regression machine parameters optimization problems are introduced in the field, such as the usage of genetic algorithms, particle swarm optimization, artificial immune algorithm, chaotic algorithms and so on. However, as the inherent shortcomings of these methods, these classical intelligent algorithms demonstrate different levels of slow search convergence speed, falling into local optimal solution easily and other related defects.In this study, the introduction of artificial immune algorithm and particle swarm hybrid method, artificial immune algorithm and chaotic local search algorithm mixed-hybrid method of these two new methods are trying to resolve the parameters optimization problem of support vector regression machine. In the immune particle swarm algorithm, by the virtue of artificial immune algorithm's easily jumping out of local minimum combined the benefits of high speed of searching features of particle swarm optimization algorithm, efficient usage of this hybrid algorithm obtains the characteristics of classical algorithms leads to much more excellent optimization ability compared to original algorithms. Chaos optimization algorithm uses the nature of chaos in order to obtain the global search ability of the optimization algorithm, by adding chaotic local search steps into artificial immune algorithm, the original algorithm obtains the nature and the ability of chaotic optimization, so the artificial immune algorithm has much stronger and more stable search capabilities than original artificial immune algorithm and chaos optimization algorithms.In the experimental part, for immune particle swarm optimization algorithm seeking support vector regression parameters, respectively, yi=sinxi+ξi and yi= sinxi/xi+ξi generated 100 two-dimensional data in the simulation data with the mixed Gaussian noise. In the actual data part, these are used in high-dimensional infrared spectroscopy of this diverse material content data and the geographical environment, natural climate data on forest fire, At the same time, with the operation of the original artificial immune algorithm and particle swarm optimization algorithm, it proves that the immune particle swarm optimization algorithm seeking the support vector regression parameters effectively access to the advantages of artificial immune algorithm and PSO, it has rapid searching speed and wide range of searching field, it has the good ability of jumping over local optimal solution, avoiding the phenomenon of premature in large extent.In the experiment of searching Support Vector Regression parameters by Chaos Immune Algorithm, using yi=sinxi/xi+ξi, mixed Gaussian noise generated by 100 two-dimensional data as the simulation data to regression, and use the compressive strength of concrete high-Victor sample data as the actual data. By operation with the original artificial immune algorithm and chaos optimization algorithm for computing at the actual same time, the comparison tests look for Chaos Immune Algorithm of Support Vector Regression parameters in these data to focus on regression problems. Experiments proved that by adding chaotic local search, the system's performance over the same time, the original artificial immune algorithm and chaotic algorithms, make the system possess better global search capability. While adding elements of chaos, the capability of jumping local optimal has been greatly enhanced.In this study, by fully understanding of support vector regression machine on the basis, and inherited the work of their predecessors, the intelligent algorithms applied to the support vector regression machine's preferences work. Through the existing principles of current intelligent optimization algorithms, the combination of these main algorithm ideas, this paper presented two innovative immune particle swarm algorithm and chaos immune algorithm used to parameter selection of support vector regression. These two algorithms are simplistic, easy to implement, less code, and the effect of good results are stable. Through theoretical analysis, and simulation and actual data validation and testing, usage of the immune particle swarm algorithm and chaos immune algorithm selection parameters obtained for support vector regression are in a relatively short time and low computational cost to find a good parameter combination. In the parameter optimization process, these two algorithms organic absorbing the advantages of the original classic algorithms, while in large extent avoiding the shortcomings of original algorithms, these two algorithm shows good capabilities of convergence speed, avoidance of premature, and support vector regression model gets high precision, excellent forecasting effect and good capability of generalization.This research introduces innovative immune particle swarm algorithm and chaos immune algorithm for support vector regression parameters selection process, it partly solved the harassment of many years of support vector regression parameters selection problem by implementation of two kinds of new and effective method compared with other commonly used methods:the ergodic method gets good results but the speed is too slow, secant method is fast but still get poor results, experience method is too unstable. These two innovative methods has a smaller computational cost getting better results,In the use of support vector regression machine to carry out regression analysis in a number of areas, support vector regression machine users are able to use a good combination of parameters to carry out regression analysis. On the other hand, the combination of a variety of intelligent algorithms used in support vector regression machine preferences, that will expand the application areas of intelligent algorithms, through the theoretical analysis and experimental results show that, by several kinds of intelligent algorithms by means of some form of organic combined, one can keep the original algorithm the merits of its own but to avoiding their own flaws.Nature is complex, with the regression analysis is widely used in finance, construction, weather and other fields, more and more high-dimensional complex non-linear regression problems need to be solved using a solid statistical theory supported by Support Vector Regression. Parameter selection is a necessary step to apply statistical learning model. The immune Particle Swarm and Chaos Immune Algorithm for Support Vector Regression parameters selection gets better balance between the run-time and operating results, these will be used more and more widely.