一种基于信念修正思想的svr增量学习算法-控制与决策.pdf

第30 卷 第7 期 控 制 与 决 策 2015 年 7 月 Vol. 30 No. 7 Control and Decision Jul. 2015 文章编号: 1001-0920 (2015) 07-1315-06 DOI: 10.13195/j.kzyjc.2014.0757 一种基于信念修正思想的SVR 增量学习算法 陈海燕, 丰文安, 王建东, 王寅同, 孙 博 (南京航空航天大学计算机科学与技术学院，南京210016) 摘 要: 针对实际应用中数据的批量到达, 以及系统的存储压力和学习效率低等问题, 提出一种基于信念修正思想的 SVR 增量学习算法. 首先从历史样本信息中提取信念集, 根据信念集和新增数据的特点选择相应的信念集建立支持 向量回归模型并进行预测; 然后对信念集进行修正, 调整当前认知状态, 使该算法对在线和批处理增量学习都有很好 的适应性. 在标准数据集上的测试验证了算法的良好性能; 在某机场噪声实测数据上的对比实验也表明, 该算法的性 能明显优于传统学习算法和一般增量学习算法. 关键词: 信念修正；增量学习；支持向量回归；认知状态；机场噪声 中图分类号: TP301.6 文献标志码: A An SVR-based incremental learning algorithm based on belief revision CHEN Hai-yan, FENG Wen-an, WANG Jian-dong, WANG Yin-tong, SUN Bo (School of Computer Science and Technology，Nanjing University of Aeronautics and Astronautics ，Nanjing 210016，China ．Correspondent ：FENG Wen-an ，E-mail ：fengwenan@126.com) Abstract: In order to reduce the storage cost and improve the learning efficiency when data arrives in batches in practical applications, an SVR-based incremental learning algorithm based on belief revision is proposed. In this algorithm, belief sets are extracted from the historical samples firstly, and appropriate belief sets are selected to establish the prediction SVR model according to the characteristics of belief sets and the new incremental data set. Then, the belief sets are revised and the current epistemic state is adjusted to adapt the new environment. The algorithm perform well when applied to online learning and incremental learning. Experiments conducted on the standard data sets and the airport noise data set demonstrate that the proposed alg