摘要
介绍了CBR与模糊结合的技术在预测领域中的应用现状,阐述了一种经作者改进的模糊k NN技术的工作原理,详细论述了时间范例相似度的匹配算法与预测算法.给出该算法在电力负荷预测中的应用的实例,描述了预测系统工作原理、系统结构设计,最后给出了实验结果.
After introducing the application status quo of CBR combined with fuzzy, this paper presents the working principle of an improved fuzzy kNN algorithm in detail. The main object of this improved method is to solve the problems of current fuzzy CBR forecasting methods. It uses linear weighted operation instead of MAXMIN operation, gradual time to forget algorithm and employs post process technology so that this method is suitable for nonstationary time series. To clarify the working principle of the method, this paper first gives the definition of the similarity of temporal cases. The similarity computing includes 3 levels: (1) the similarity computing on case's attributes; (2) the similarity computing on case points; (3) the similarity computing on temporal cases. Then the paper gives the matching and forecasting algorithm of temporal case, presents the import data structure of the algorithm. The main forecasting process includes 3 steps: (1) Construct a case point at 0th time by getting the time series data at 0th time, and then form a temporal case by combining the 0th case point and a sequence of past case points. (2) Search k cases which are most similar to the temporal case by using fuzzy kNN technology. (3)Compute the future trend according to the similar cases.To demonstrate the effects of the improved forecasting technology, this paper gives a experimental result on power load forecasting. The experimental result shows that the short period forecasting precision is better than that of human beings. Employing this method can avoid constructing and tunneling the model of power load, which is very time consuming.
出处
《南京大学学报（自然科学版）》
CAS
CSCD
北大核心
2003年第2期159-164,共6页
Journal of Nanjing University（Natural Science）
基金
国家自然科学基金(60075015)
