面向配电网故障数据的BIC评估后向选择方法

10 kV配电网所处环境复杂,引发故障的原因很多,在使用数据挖掘方法对配电网故障进行分析时,太多的特征会对挖掘模型造成负面影响.为了防止挖掘模型考虑过多无用信息,需首先对数据进行特征选择来实现降维,因此提出了基于贝叶斯信息准则(BIC)的模型评估后向选择算法,对故障因素进行降维.BIC评估准则能够尽可能地简化模型,降低维度,而后向选择算法可以快速得到最优的简化模型,两者的结合提升了降维的速度,并能够得到更加简化的模型.实验结果表明,采用基于BIC评估的后向选择算法有助于后续模型准确性的提升,可提高训练效率.

An BIC Selection Method for Distribution Network Fault Data Feature Dimension Reduction

Feature selection is important to improve the model accuracy and reduce overfitting.The 10 kV power distribution network is comnplex and there are too many features for a data mining model to work.Before modeling power fault data,the dimensionality reduction and model selection is necessary.In order to solve this problem,a Bayesian information criterions (BIC) model selection algorithm along with backward selection algorithm was proposed.BIC aims to reduce the complexity of model and the backward selection can reach fast convergence.Experiments show that the algorithm works well.It is proven that the algorithm proposed here is of advantage to improve model accuracy and data training efficiency.