基于神经网络的日峰荷预测方法中日期类型系数的确定（英文）

Weekday Index Determination for ANN-Based Daily Peak Load Forecasting Method

对于基于人工神经网络的短期负荷预测来说,日期类型(星期几)是需要考虑的重要影响因素。通常,日期类型系数被编成7位二进制码作为神经网络的输入变量。该文提出一种日期类型系数的确定方法,将日期类型系数编为1位输入变量,由于精简了输入量,从而提高了预测精度。该日期类型系数通过计算不同日期类型的负荷–气温散点图的拟合曲线、并估计不同日期类型的负荷之差得到。为了消除夏季气温累计效应对负荷的影响并得到更清晰的负荷日期特征,该文采用一种基于遗传算法的气温修正方法,对气温变量进行修正。最后通过对苏州的实际负荷数据预测验证了方法的有效性。

For the ANN-based short-term load forecasting, the weekday index（day of week） is a very important influencing factor that needs to be considered. Usually, the weekday index is coded as 7 binary inputs. This paper proposed a method that codes the weekday index as only one input variable. Due to the simplicity, the precision of the forecasting was improved. The proposed weekday index is determined by estimating the load＇s difference of different day types. For this purpose, the regression curves of load versus temperature of different day types were calculated. To eliminate the accumulation effect and get a clearer weekly pattern of these regression curves, a genetic-algorithm（GA） based method was adopted to modify the temperature variables. The proposed method was demonstrated by the calculated results on the load data of the Suzhou city in China.