电力系统负荷预报的神经网络LBP算法

LBP ALGORITHM OF NEURAL NETWORK FOR POWER LOAD FORECASTING

本文提出了一种能够反映工作日电力负荷波动性并可同时进行假日负荷预报的神经网络算法。该算法在一个神经网络中构造多个相互关联的子网络，将一周７日根据负荷特点分为四类特征日期，通过解码器根据输入的日期特征量激活对应的子网络，对其训练并作出预报。通过对实际系统的实验表明，该算法具有较高的预报精度。

A new kind of neural network algorithm for hourly load forecasting, consisting of several sub neural networks in structure, is introduced in this paper. For the different characters of hourly load on every weekday, the seven days of a week are divided into four types. After the patterns of every typical day have been taken for training every sub neural network, the network can be used to forecast the hourly load of different typical days by different sub neural networks. A practical example shows that this algorithm can reflect the influences of atmospheric temperature and day type on power load.