论文部分内容阅读
针对支持向量机(SVM)最佳算法参数难以确定的不足,以及标准遗传算法(SGA)易陷入局部极值和早熟收敛等问题,利用量子遗传算法(QGA)搜寻SVM学习参数——惩罚因子C、核函数参数g和不敏感系数ε,提出2种不同搜索空间方案的QGA-SVM预测模型,并构建SGA-SVM模型作为对比。以云南省上果站年径流预测为例进行实例研究,利用实例前30年和后16年资料对模型进行训练和预测,结果表明:①2种方案中QGA-SVM模型对实例后16年年径流预测的平均相对误差绝对值分别为3.02%、2.77%(5次平均),精度优于SGA-SVM模型,表明QGA-SVM模型具有较高的预测精度和泛化能力;②QGA采用基于量子位的编码和基于量子门的种群更新策略,使得该算法隐含强大的并行性并能更好地维持种群多样性,即使在更大的搜索空间也能获得较好的寻优效果。该算法具有种群规模小、收敛速度快和全局寻优能力强的特点,利用QGA算法优化得到的SVM学习参数有利于提高SVM模型的预测精度和泛化能力。
In order to solve the problem of difficult to determine the optimal algorithm parameters of support vector machine (SVM), and the problem that standard genetic algorithm (SGA) is easily trapped in local extremum and premature convergence, QGA is used to search SVM learning parameters - penalty factor C , Kernel function parameter g and insensitivity coefficient ε, two QGA-SVM prediction models of different search space schemes are proposed, and the SGA-SVM model is constructed as a comparison. Taking the annual runoff forecast of Shanggou Station in Yunnan Province as an example, the model was trained and predicted by using the data of the first 30 years and the last 16 years of the example. The results show that: (1) The QGA-SVM model of two kinds of scenarios The average relative error of prediction is 3.02% and 2.77% respectively (5 averages), and the accuracy is better than SGA-SVM model, which indicates that QGA-SVM model has higher prediction accuracy and generalization ability. Coding and population-based population update strategy, making the algorithm implies a strong parallelism and can better maintain the diversity of populations, even in a larger search space can get a better search results. The proposed algorithm has the characteristics of small population size, fast convergence rate and global optimization ability. The SVM learning parameters optimized by QGA algorithm are helpful to improve the prediction accuracy and generalization ability of SVM model.