基于人工神经网络的电力负荷坏数据辨识与调整

电力负荷坏数据辨识应充分考虑负荷曲线本身的特征。先用Kohonen网对日负荷曲线进行聚类，产生各类的特征曲线；然后用特征曲线及由此产生的含有坏数据的曲线形成的样本集对BP网进行训练，利用BP网的泛化能力，使之具备对本类曲线进行坏数据精确定位的能力；最后利用特征曲线进行坏数据的调整。该方法能够做到离线训练，在线辨识，实例分析取得了良好的效果。

OUTLIER IDENTIFICATION AND JUSTIFICATION BASED ON NEURAL NETWORK

A new method is presented to identify outliers in load data by fully utilizing the features of electrical load curves. First, the day load curves are clustered by a Kohonen neural network, and a typical load curve is thus obtained for each cluster. Then a BP neural network is trained with each typical load curve and some other curves derived from it with some outliers included. Owing to its generalization ability, the network can identify the outliers in the curves included in the corresponding cluster. At last, the outliers are adjusted with typical curves. The off line trained neural networks can be used to identify the outliers on line. Test results using actual data are served for demonstrating the feasibility of the proposed method.