【摘要】 针对训练样本与负荷预测模型的构建及预测精度之间的强相关性,在对负荷变化规律深入研究的基础上,提出了样本动态组织理论与方法.根据负荷变化的横向与纵向特征、日期、季节特征和气象特征构建时间分类树和样本映射表,并通过对气象数据的模糊化处理进行样本初选,进而利用自组织网络(SOFM)的改进方法提取负荷水平变化趋势的特征曲线,以实现样本的动态精选.多种模型的预测结果表明,采用的由粗到精逐步细化,多层面、多角度的样本过滤机制,为预测日负荷建模提供了更加优质的历史样本,很好地抑制了不良样本对预测建模可能带来的各种干扰,有效提高了电力系统短期负荷预测精度.更多还原

【Abstract】 For the strong correlations among training examples,load forecasting model construction and forecasting accuracy,on the basis of researching into the load changing law,the theory and corresponding method for dynamic example organization were presented.According to the horizontal and longitudinal load changing characters,date,season and weather characters,the time classification tree and example map were constructed.Then,the examples were selected preliminarily via fuzzily dealing with weather data,and the dynamic selection of examples was completed via extracting the characteristic curve of load level changing trend by the improved self-organizing feature map(SOFM).The results of several forecasting models indicate the filtration mechanism with several levels and perspectives can offer better training examples to inhibit interference from bad examples and get higher forecasting accuracy of short-term load.更多还原