欢迎您！东篱公司
退出
-
申报数据库
-
立项数据库
-
成果数据库
-
项目获奖数据库
中文摘要：
利用数据挖掘中的聚类技术将历史负荷数据进行聚类．根据聚类后的分类信息对定性属性利用粗糙集进行属性约简，由约简结果进一步生成决策规则树供短期预测使用。根据聚类的结果对每类进行BP神经网络的训练，神经网络的隐含层单元采取逐步试用的方法根据训练误差最小化进行选择。在实际预测中，首先对待预测的记录利用决策规则树进行归类．然后选取相应类别的神经网络予以预测。通过实例证明，该方法的平均相对误差为2．16％．而同结构BP神经网络预测的平均相对误差为2．67％．ARMA预测的平均相对误差为3．81％．证明所提方法有效。
英文摘要：
The clustering technology of data mining is used to classify the historical toad data and the rough set is then used to reduce the related qualitative attributes of the clustered information,based on which,a decision tree is constructed to provide the rules for short-term load forecasting. The historical data of every class are used to train the BP-ANN network,the number of hidden layer nerve cells increases step by step in an experiential range until the minimal train error is achieved. The class of data for forecasting is decided by its qualitative factors according to the rules and the BP-ANN for forecasting is accordingly chosen. Example shows that, the mean relative error of and ARMA are 2.67 % and 3.81% respectively. proposed method is 2.16 %,while those of BP-ANN
同期刊论文项目
同项目期刊论文
Multi-objective coordination optimal model for new power intelligence center based on hybrid algorit
Electricity price forecasting using generalized regression neural network based on principal compone
期刊信息
位置：