神经网络短期负荷预测输入变量选择新方法

输入变量选择是神经网络建模前的一项重要工作，是否能够选择出一组最能反映期望输出变化原因的输入变量直接关系到神经网络预测的性能。文中将正交最小二乘(OLS)法应用于神经网络短期负荷预测的输入变量选择。以南京地区1998年、1999年夏季日最大负荷预测为例，对比了OLS法与相关系数法的输入变量选择结果。结果显示OLS法可以得到更小、更准确的输入变量集，神经网络的收敛速度更快，预测结果也更好，从而验证了该方法的有效性。

NOVEL INPUT VARIABLE SELECTION FOR ANN SHORT-TERM LOAD FORECASTING

It is im portant to selectinput variables for ANN based load forecasting.Whether the input variables,representing the cause of change of expected output,are selected or not is relevant to the perform ance of ANN forecasting. A novel method of input variable selection for ANN short- term load forecasting based on OL S m ethod is presented.An example of forecasting Nanjing 1998and 1999summer day peak load is given and the results obtained by both OL S method and correlation coefficient analysis method are compared.The comparison shows that smaller and more accurate input variables set can be obtained via OL S method,thus its effectiveness is demonstrated.