Abstract: An ultra-short term load forecasting method based on robust regression and echo state network is built considering the very short forecast time of ultra-short term load forecasting. As a recurrent neural network, echo state network has a dynamic reservoir as the hidden layer and trained by linear regression, which can model any given function and make it training very quickly. As a result, echo state network can meet the needs of ultra-short term load forecasting. The using of robust regression to training can effectively reduce the effects of outliers and promote the accuracy of forecasting. The results of calculation example show that the proposed method is feasible and effective.

Key words: ultra-short term load forecasting, echo state networks, robust regression, power system

摘要： 针对超短期负荷预测周期短，要求预测速度快的特点，构建了基于稳健回归和回声状态网络的超短期负荷预测方法。回声状态网络作为一种递归神经网络，其隐含层为一个储备池，并且通过线性回归训练网络，从而具有映射复杂动态系统的能力和训练快速的特点，能较好地满足超短期负荷预测的要求。考虑到异常负荷数据的影响，将稳健回归运用于网络训练阶段，以削弱异常值的影响，从而提升预测的精度。通过算例验证了所提方法的可行性和有效性。

关键词: 超短期负荷预测, 回声状态网络, 稳健回归, 电力系统